Arkham now puts full wallet history, cross-chain tracing and risk scores behind a paid Enterprise tier. I followed a real $26 million theft with two Bitquery tools, then lined up five more options.

Arkham’s free plan now shows only the last 100 transactions of any wallet. To see the rest, you need Arkham Enterprise. That paid tier also holds risk scores, case files and tracing from one chain to another. Arkham launched it on 20 October 2026, and its announcement lists no price.

A busy wallet can make 100 transactions in an afternoon. So if you chase stolen funds or check wallets for a compliance team, the free plan now stops early. If that is you, you need an Arkham Intelligence alternative, at least for part of the work.

I wanted a fair test, so I took one real theft and followed it to the end. On 8 January 2026, an attacker took 8,535 ETH from Truebit, an Ethereum project with its own token. That was about $26.4 million at the time. Within 53 hours, all of it had gone into Tornado Cash, a mixer that pools many people’s coins so the trail breaks.

I ran that trail through both Bitquery tools myself. For the other five, I say what each does best and where it stops.

The short version

• Bitquery MoneyFlow (lite.bitquery.io): a free, Arkham-style flow chart in your browser. Paste a wallet, pick one of 17 chains, and follow the money one step at a time.
• Bitquery MCP: for people who work in Claude, ChatGPT or Cursor. Ask in plain English, and the AI pulls live transfers and labels from 19 chains, plus risk scores for Ethereum, Bitcoin and TRON.
• Nansen: for trading research and “smart money” wallets.
• Chainalysis Reactor and TRM Labs: what police and large compliance teams buy. Prices come by quote.
• Dune: for people who write SQL and want dashboards.
• Etherscan: free, and fine for checking one wallet or one transfer at a time.

Why people look for an Arkham Intelligence alternative

Arkham Intelligence is a blockchain analytics platform, and it built its name on labels. A label is a name attached to a wallet address, such as “Binance hot wallet”. Arkham says it holds 3.9 billion “attributions”, its word for these labels. Those labels stay free, along with its graph tool, called Visualizer, and its block explorer.

Here is what sits behind Arkham Enterprise on 20 October:

• the full transaction history of a wallet, with filters (free users see the 100 most recent transactions)
• Tracer 2.0, which follows money across bridges, the services that move coins between chains
• the counterparties panel, which shows who a wallet deals with most
• risk scores and case management
• creating new custom labels

Arkham does give free Enterprise access to some independent investigators and public-interest groups through a grant program. Everyone else has to ask for a price.

The API is gated too, with one exception. You apply through a form, and once Arkham approves you, paid plans start at $100, billed in monthly credits. Arkham also hosts an MCP server, so Claude, Cursor or Codex can query its data, but it needs that API plan or a trial. The exception is for AI agents: about 90 read endpoints sold one call at a time, at $0.20 per credit in USDC, with no account.

I get why Arkham moved full history behind a paywall. Serving full history for every wallet costs money. Still, Arkham keeps one big edge. If your main question is who owns a wallet, its free labels are hard to beat, and nothing on this list replaces them one for one. The tools below compete on history, access and price instead.

The test case: the $26 million Truebit hack

Truebit sold its TRU token through its own sale program on Ethereum, written in 2021. A math bug in that old code made the price of new TRU come out as zero. The attacker minted TRU for free, sold it back for ETH, and repeated the trick several times in one transaction. Security firm SlowMist put the loss at $26.44 million.

Here is the trail, as the two Bitquery tools showed it:

1. 8 January, 16:02 UTC. 8,535 ETH lands in the attacker’s wallet, 0x6C8E鈥�2b50 (transaction).
2. Same day. The wallet splits the money: 4,267 ETH to one new wallet, 4,001 ETH to another. A third payment of 268 ETH follows on 10 January.
3. 8 to 10 January. The 4,267 ETH pile hops through three more new wallets. The 4,001 ETH pile makes one hop.
4. 10 January. Two final wallets feed Tornado Cash in round amounts of 100, 10, 1 and 0.1 ETH. By 20:52 UTC, 8,535 ETH has gone in, the same amount that was stolen.

Any tracing tool should answer three questions about a trail like this. Where did the money come from? How did it split? Where did it end up?

1. Bitquery MoneyFlow: visual crypto tracing in your browser

MoneyFlow is Bitquery’s browser tool for tracing money between wallets. It lives at lite.bitquery.io, and it is free: you only need to sign in with a Bitquery account.

Starting a trace takes four steps:

1. Paste an address.
2. Pick the chain. On 29 October the menu listed 17: Bitcoin, Ethereum, BNB Smart Chain, Polygon, Arbitrum, Base, Optimism, TRON, XRP Ledger, Cardano, Litecoin, Dogecoin, Dash, Algorand, Filecoin, Arc and Robinhood Chain.
3. Pick the coin or token. MoneyFlow lists every currency the wallet has touched, with the date it last moved.
4. Set a date range if you want one, then press Create MoneyFlow.

The result is a flow chart. Money coming in sits on the left, money going out sits on the right, and the width of each band shows the amount. For the Truebit wallet, one green band of 8,535 ETH comes in and splits into red bands of 4,267, 4,001 and 268 ETH.

Trace Out adds the next step of the trail, and Trace In walks back toward the source. Each step saves a new version of the trace, so you can go back if a step gets messy.

The table under the chart shows Bitquery’s labels next to each address. Both wallets that took the big piles carry the tag hacker-truebit-hacker. Exchange wallets get tagged too. When I traced the Bybit hacker’s wallet from February 2025 the same way, the money came from a Bybit wallet tagged bybit-proof-of-reserves.

The chart also showed something I did not expect. A fan of tiny payments, each worth less than $1, came into the attacker’s wallet from look-alike addresses. One look-alike copies the first and last four characters of a wallet the attacker used, so both read 0x3b58鈥�862a in short form. The trick is called address poisoning. Scammers hope you copy the fake address from your history next time, and here they aimed it at the thief. A threshold box above the chart hides payments below any dollar amount you choose.

Two more buttons help when you need to share the work. Generate report turns the trace into a table of every transfer, with full hashes, which you can download as a CSV file. Get API opens the query behind the chart in Bitquery’s GraphQL IDE, so a developer can run it again.

Where it falls short. Solana is not in the chain menu yet. Each trace follows one currency, so if a thief swaps ETH for USDT, you start a second trace. And after a few steps the chart can turn into loops when money flows back into wallets already on it. In the Truebit case, the third step got hard to read, and the table was clearer.

2. Bitquery MCP: ask your AI assistant to follow the money

MCP, short for Model Context Protocol, is a standard way to plug outside tools into an AI assistant. Bitquery MCP plugs live blockchain data into Claude, ChatGPT, Cursor, VS Code and OpenAI Codex.

Setup is short. Add https://mcp.bitquery.io as a custom connector in your assistant, then sign in with your Bitquery account. There is no API key to paste.

You then get 314 tools. They trace transfers on 19 chains, including Solana, Bitcoin, Ethereum, TRON, the XRP Ledger and Stellar. They also cover DEX trades and prices on 10 chains. Guided prompts handle the common jobs: money_flow follows funds until they reach an exchange or a mixer, and risk_screen returns a LOW, MEDIUM or HIGH verdict with evidence.

I gave Claude the Truebit attacker’s wallet and asked where the ETH ended up. I also asked it to name every wallet on the way and score the last ones for risk. It pulled the transfers hop by hop, looked up the labels and ran a risk check on each final wallet.

The answer named two exit wallets. The first, 0x7720鈥59d, got 4,267 ETH after three more hops and sent 4,266.9 ETH to Tornado Cash on 10 January. The second, 0xd841鈥135, got the 4,001 ETH pile plus the 268 ETH payment and sent 4,268.5 ETH to Tornado Cash the same evening. Both scored 100 out of 100 for risk, the top of the scale, mainly because they paid a mixer. The risk check also flagged look-alike poisoning addresses next to both.

MoneyFlow reached the same Tornado Cash address three steps out from the attacker. The MCP got there from one question. Arkham has a hosted connector too, so the real difference is access. Arkham’s needs an approved API plan or trial, while Bitquery’s trial is self-serve and needs no card.

The product page listed three plans on 29 October 2026. The free trial runs 7 days with 100 credits, and it needs no card. Trading MCP costs $19 per month and covers trading data only, with 32 tools and no tracing. AI Investigation MCP, the plan with tracing, costs $149 per month, or $100 per month billed yearly, and includes all 314 tools.

Where it falls short. On Ethereum the risk check looks one step out from a wallet; on TRON it can look several steps. An AI assistant can also misread a result, so open the transaction hashes it cites before you rely on them. And like MoneyFlow, it follows one chain at a time, so you hop a bridge yourself.

Arkham vs Bitquery at a glance

In words: Arkham’s free plan shows the last 100 transactions per wallet, while MoneyFlow uses whatever date range you set and the MCP pages through the full history. Arkham’s Visualizer stays free, and MoneyFlow draws a flow chart too; the MCP answers in text.

Arkham Enterprise follows money across bridges for you. MoneyFlow and the MCP work one chain at a time. Risk scores come with Arkham Enterprise and with Bitquery MCP, not with MoneyFlow.

Price works differently on each side. MoneyFlow is free with a Bitquery account. Arkham Enterprise lists no price. Its API starts at $100 after approval, or $0.20 per credit pay as you go. Bitquery MCP is self-serve, with a free trial and then $149 per month for tracing.

5 more Arkham alternatives worth knowing

3. Chainalysis Reactor

Best for police, prosecutors and large exchanges. Reactor is Chainalysis’s case graph, and law enforcement uses it widely. There is no public price and no self-serve trial. If you work alone or on a small budget, it is likely out of reach.

4. TRM Labs

Best for compliance teams and government buyers. TRM sells its Forensics tool for tracing funds across chains, next to its wallet screening tools. Most of it is sold by quote, mainly to large firms.

5. Dune

Best for analysts who write SQL. Dune lets you query blockchain data and build dashboards, and it has a free tier and its own MCP server. There is no ready-made tracing screen, so you write the queries yourself or let its MCP server write them.

6. Etherscan

Best for checking one transaction or one wallet for free. Etherscan shows name tags on well known addresses, and it tags the Truebit attacker’s wallet “Truebit Exploiter 1”. There is no flow chart, so tracing a trail across many hops means a lot of tabs.

7. Nansen

Best for trading research. Nansen labels wallets and marks the ones with a strong record as “smart money”. You can then watch what they buy and sell. There is a free tier and a paid Pro plan, plus an API, an MCP server and a command-line tool. Its tools center on trading signals, so it is a weaker fit if you need to build a case file.

Other Bitquery products for investigators

Bitquery sells three more things for this kind of work:

• Coinpath: the money-flow data behind MoneyFlow, sold to companies with full history and bulk export.
• Crypto investigation service: if you were robbed and would rather hand the work to someone, Bitquery’s team traces the funds. Its page says the first assessment is free.
• Ethereum datasets: the full history of the chain as files, for teams that want to run their own analysis.

How to choose an Arkham alternative

• You want Arkham’s look without paying for Enterprise: Bitquery MoneyFlow, which is free.
• You already work in Claude or ChatGPT: Bitquery MCP.
• You trade and want smart-money signals: Nansen.
• You work for police or a regulated firm with a budget: Chainalysis Reactor or TRM Labs.
• You write SQL: Dune.
• You only need to check one transfer: Etherscan.

FAQ

Is there a free alternative to Arkham Intelligence?

Yes. Bitquery MoneyFlow is free with a Bitquery account and traces a wallet over any dates you pick. Etherscan is free for looking up one wallet or transaction at a time, and Dune has a free tier. Bitquery MCP has a 7-day free trial with 100 credits and no card. Arkham itself still has a free plan, but it shows only the last 100 transactions per wallet.

What is the best Arkham alternative for tracing stolen crypto?

For a visual trail, Bitquery MoneyFlow; for asking questions in plain English, Bitquery MCP. For court work with a large budget, Chainalysis Reactor or TRM Labs. I work at Bitquery, so weigh my pick with that in mind.

Does Arkham have an API?

Yes. You apply through Arkham’s API form, and once Arkham approves you, plans start at $100, billed in monthly credits. Some applicants get trial access first. Arkham also hosts an MCP server for AI assistants, and AI agents can pay per call at $0.20 per credit in USDC.

Can AI trace crypto transactions?

Yes, if the assistant is connected to live blockchain data. With Bitquery MCP, Claude followed the Truebit money to Tornado Cash and scored both exit wallets 100 out of 100 for risk. Check the transaction hashes it cites before you act on them.

Can you trace crypto after it goes into Tornado Cash?

Not directly. A mixer pools many people’s deposits, so the link between a deposit and a withdrawal breaks. Investigators match the timing and size of later withdrawals, which is slow and gives odds, not proof.

Try it yourself

Start a trace at lite.bitquery.io, or connect Bitquery MCP to your AI assistant and ask it to follow a wallet. If you want the raw history instead, Bitquery sells the Ethereum dataset as files.

Best Arkham Intelligence Alternatives in 2026: 7 Tools, 2 Tested on a $26M Hack was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

Danelfin and Tickeron both sell “AI” to people picking stocks, but they are not the same kind of product. Danelfin gives every stock a score from 1 to 10 and leaves the trading to you. Tickeron sells signals and trading bots, some of them running on its own brokerage accounts. I read both sites end to end on Oct 26, 2026, including the parts most reviews skip: Danelfin’s own audit site and the small print under Tickeron’s 191% headline.

I also work on milo, so I’ll say where milo fits and where it doesn’t. It isn’t a third option that beats both. It answers a different question.

The short answer

Pick Danelfin if you want a daily 1 to 10 rating on about 5,600 US and European stocks and you’ll place the trades yourself. Pick Tickeron if you want ready-made signals or bots that trade on 5 to 60 minute timeframes. Both lead with backtests. Danelfin publishes an audit that shows its live record so far is flat. Tickeron’s headline numbers don’t come with dates.

Danelfin in plain English

Danelfin calls itself “AI-Powered Stock Picking.” Every trading day it gives each stock an AI Score from 1 to 10, which it describes as the probability of beating the market over the next three months. A 10 means the model likes it most, a 1 means least. You also get sub-scores for technical, fundamental, sentiment and low risk.

How the AI Score works

Per Danelfin’s how-it-works page, the model looks at more than 600 technical, 150 fundamental and 150 sentiment indicators a day, which it turns into 10,000+ features per stock. It uses hundreds of decision trees and trains on the latest 252 market days for US stocks. Coverage is all US stocks, the main European markets and US-listed ETFs. The audit site puts the count at roughly 5,600 stocks scored daily.

Danelfin pitches this as explainable AI, “no black boxes,” because it shows which features pushed a score up or down. That’s useful. It still won’t tell you why a 9 should beat an 8 in your account.

One thing to get straight early: Danelfin doesn’t trade for you. The audit site says it plainly, “you execute the trades on your own brokerage account.” The paid plans can sync a US broker portfolio so you see scores on what you own, but that’s a read, not an order.

Danelfin pricing

From danelfin.com/pricing on Oct 26, 2026:

• Free: $0. Top 10 lists, 10 stock reports a month, one portfolio of 5 stocks, top alpha signals only.
• Plus: $22 a month billed yearly ($264). Adds portfolio sync with US broker accounts.
• Pro: $59 a month billed yearly ($708). Adds full alpha signals and CSV export.
• Elite: $134 a month billed yearly ($1,608). Adds an API with 10,000 calls a month and historical daily scores back to 2017 on request.

The per-month prices are the yearly plan divided by 12. The page says yearly saves $84, $240 and $540 a year, which works out to about $29, $79 and $179 a month if you pay monthly. There’s a 30-day money-back guarantee.

The performance claims, and Danelfin’s own audit

The homepage says stocks with a Score of 10 beat the market by +21.05% on average after three months (annualized alpha) since 2017, and stocks scored 1 lagged by 33.28%. It also says its Best Stocks strategy returned +456% from Mar 29, 2017 to Sep 3, 2026, against +228% for the S&P 500.

The how-it-works page is honest about what those are: “This Danelfin Best Stocks strategy performance is based on backtested results. Backtested performance is not an indicator of actual future results.” The +21.05% figure covers Jan 3, 2017 to Apr 1, 2025.

Credit where it’s due, Danelfin runs a public audit site and the audit is tougher on Danelfin than the homepage is. Here’s what it says:

• The current model went live on Jul 7, 2025. Every score before that date was recalculated afterward, walk-forward. The audit calls this a “simulated backtest” and notes the model was picked after its performance was known.
• It splits the Score 10 backtest into three periods. Average 3-month alpha was +5.4% in 2017 to 2019, +0.3% in 2020 to 2022, and -0.5% from 2023 to Jul 6, 2025. The audit says only 2017 to 2019 is statistically significant and tells readers to “treat the recent period as unproven.”
• Live, stocks scored 8 to 10 averaged -0.5% 3-month alpha since Jul 7, 2025, against +2.2% for the same bucket in the backtest. That’s about four independent 3-month windows, so it’s early either way.
• It passed 7 of 14 robustness checks, with 6 cautions and 1 fail. The fail is t-stat inflation: naive t-stats overstate significance by about 4.9x.
• Survivorship bias isn’t controlled. Delisted companies like Twitter, SVB and Celgene are missing from the history.
• A Top 20 portfolio backtest returned 22.4% a year vs 15.1% for SPY from Apr 2017 to Jul 2026, before trading costs, with about 70% turnover at each monthly rebalance.

Six-month alpha tells the same story: +4.8%, then -1.4%, then -2.6%. So the big homepage number leans mostly on 2017 to 2019. That doesn’t make the score useless as a screen. It does mean nobody, Danelfin included, can show it beats the market now.

Tickeron in plain English

Tickeron’s homepage reads “AI Trading Agents, Tools and News.” Where Danelfin is one score, Tickeron is a shelf of products: Daily Buy/Sell Signals, AI Robots, Real-Time Patterns, Trend Prediction, Pattern Search and a screener, plus crypto signal agents. Its models, which it calls Financial Learning Models (FLMs), run on 5-minute, 15-minute and 60-minute charts. That’s short-term trading, not “hold for three months.”

Bots, signals and patterns

The bot trading page talks about “1,000+ back-tested algorithms, verified forward testing” and lists 25 bots selected from 351. The homepage the same day said 30 selected out of 298. The bot count moves, which makes sense if bots get added and retired. Some are “Double Agent” bots that hedge with inverse ETFs, for example pairing $NVDA with NVDS or $MSFT with QID.

Brokerage Agents vs virtual agents

This is the part people mix up. Tickeron’s AI agents page says Brokerage Agents are “designed for those who intend to use them for trading on real-money accounts,” and invites you to “review trades and copy from our brokerage accounts.” So Tickeron runs the bot on its own account and you mirror the trades. Virtual agents and signal agents don’t trade real money at all.

Can a bot place orders straight into your own broker account? Tickeron’s FAQ, dated Nov 22, 2023, said robot brokerage integration was “under development” with plans for Binance, Alpaca, Kraken, Ameritrade and IBKR. I couldn’t find a current page confirming it shipped, so ask them before you pay for that.

Tickeron pricing

There’s no public pricing page. Both tickeron.com/pricing and /pricing/ returned 404 when I checked. The only prices I found were in a Sep 21, 2026 sale post: Daily Buy/Sell Signals at $60 a year instead of $240, and AI Robots at 50% off without a dollar figure. Expect to see prices after signing up.

The 191% claim

The bot page headline is “191% Annualized Return During the Reported Period,” with 419 closed trades, a 71% win rate, a 2.92 profit factor, $456 average profit per trade and a 13:1 profit-to-drawdown ratio. It adds “Past performance does not guarantee future results.”

What the block doesn’t say is which bot that is, what dates “the reported period” covers, whether it was live money, paper or a backtest, and what a benchmark did over the same window. 419 trades is a decent count, but annualized returns from a short window can look huge. Until you know the dates, 191% is a hard number to judge.

Danelfin vs Tickeron side by side

Where milo fits, and where it doesn’t

milo doesn’t score stocks and doesn’t run stock bots. On the stock side it reads filings. The andmilo.com/stocks page pulls Form 4 insider buys and sales and 8-K filings from SEC EDGAR for 50 of the most-followed US stocks, refreshed daily. As of Oct 25, 2026, 23:23 UTC, it showed at least 278 insider sales worth at least $3.3B over 90 days, against at least 7 buys worth at least $54M.

Here’s a real one. On Sep 29, 2026, Oracle director Stephen Rusckowski bought 25,000 shares of $ORCL at a weighted average of $139.352, about $3.48M, filed Oct 1. I opened the Form 4 to check: code P, open-market fills between $138.96 and $139.74, held through his living trust, and the 10b5–1 plan box isn’t checked. That’s a clean buy with his own money.

Now a messy one. The same feed lists a $24M $PEP “buy.” The footnotes show it was a swap between Pohlad family trusts, not someone betting on PepsiCo. A score wouldn’t catch that and neither does a raw count. You have to read the filing, which is how to read SEC Form 4 is about.

Where milo is weaker than these two, plainly:

• No AI score or ranking of stocks.
• No ETFs or European stocks. Insider data covers 503 US tickers, against about 5,600 for Danelfin.
• No backtest or stock-picking track record to show you.
• No real-money stock trading. Stock features are paper trading and research only.
• No intraday pattern tools like Tickeron’s.

Where it’s different: every data point links back to a public SEC filing you can open yourself, and on the crypto side milo agents trade from your own wallet with public PnL and a plain-English diary of each move.

If you want the wider field, the best AI stock pickers roundup compares more tools, and can ChatGPT pick stocks covers the public AI paper-trading tests.

How to judge any AI tool’s track record

The same five questions work for Danelfin, Tickeron, milo or whatever comes next:

1. When did the model go live? Anything before that date is a backtest, however it’s labeled. Danelfin gives you the date (Jul 7, 2025). Tickeron’s 191% doesn’t.
2. Is the result split by period? One big number from 2017 can hide a flat last three years. Danelfin’s audit shows exactly that.
3. Are dead companies in the data? Without delisted stocks, backtests look better than reality.
4. Is it before or after costs? 70% monthly turnover or hundreds of short-term trades cost real money.
5. What’s the benchmark over the same dates? A return with no comparison isn’t a result.

If a vendor can’t answer those, treat its performance number as marketing. If you want to test ideas without risking money, paper trade them first. That’s the only way milo handles stocks anyway.

FAQ

Is Danelfin better than Tickeron?
They do different jobs. Danelfin is a 3-month stock rating you act on yourself. Tickeron is short-term signals and bots. Danelfin is more open about its record, since it publishes an audit with live results.

Is Danelfin’s AI Score accurate?
By its own audit, Score 10 stocks beat the market in 2017 to 2019 and were roughly flat after that. Live since Jul 7, 2025, stocks scored 8 to 10 averaged -0.5% 3-month alpha.

How much does Danelfin cost?
There’s a free plan. Paid plans were $22, $59 and $134 a month billed yearly on Oct 26, 2026, with higher monthly-billing prices.

How much does Tickeron cost?
Tickeron doesn’t publish a pricing page. A Sep 2026 sale listed Daily Buy/Sell Signals at $60 a year, down from $240.

Do Tickeron bots trade real money?
Brokerage Agents trade on Tickeron’s own real-money accounts and you can copy the trades. Virtual and signal agents don’t trade real money.

Is Tickeron’s 191% return real?
It’s Tickeron’s claim for “the reported period.” The page doesn’t name the bot or the dates, so you can’t check it from that page.

Does Danelfin place trades for me?
No. You place trades in your own brokerage account. Paid plans can sync a US broker portfolio to show scores.

Is milo an alternative to Danelfin or Tickeron?
Not a direct one. milo tracks SEC filings and insider trades for US stocks with paper trading only, and runs AI trading agents for crypto.

Try milo to see insider buys and 8-Ks for the stocks you follow, and paper trade the ideas before any money is involved.

For education only. Not investment advice. milo stock features are paper trading and research only. Competitor details checked on their own sites on Oct 26, 2026 and may change. Sources: danelfin.com (home, how-it-works, pricing), audit.danelfin.com, tickeron.com (home, bot-trading, ai-agents, FAQ, Sep 21, 2026 sale post), andmilo.com/stocks, SEC EDGAR Form 4 accession 0001341439–26–000099.

Originally published at https://blog.pumpparade.com on October 26, 2026.

I read Danelfin’s audit and Tickeron’s small print. Only one showed a dated live record. was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

AI already runs the monitoring, the simulations, and a lot of the automation behind on-chain risk. Here is what it still can’t do on its own, and why your yield still rests on human judgment.

Here is the short answer. AI already manages real parts of on-chain risk today.

It runs simulations, watches markets every second of every day, and adjusts protocol settings inside limits that people approve.

What it does not do yet is run the whole risk desk on its own. So the honest take on AI risk management in DeFi is this: yes for the hard, repetitive parts, and not yet as a manager you leave alone.

That gap matters more than the hype suggests, so let me walk through what AI does, where it hits a wall, and what measured on-chain risk really looks like.

The quick version

• AI already runs simulation, monitoring, and bounded automation across DeFi.
• Firms like Gauntlet and Chaos Labs provide much of that risk plumbing.
• Automated systems act only within human-set, governance-approved limits.
• The real limits stay: stale data, models that get gamed, and crises no model has seen.
• Over-automation can spread risk, not contain it. AI augments risk management; it does not replace it.

What “managing on-chain risk” actually means

Managing on-chain risk means keeping a protocol solvent and liquid while markets move.

In plain terms, that is four jobs. Set the rules: collateral ratios, borrow caps, interest curves. Watch everything: positions, liquidity, oracle prices, concentration.

Stress-test: model crashes and attacks before they happen. Act fast: change settings the moment conditions change.

It is constant, data-heavy work, and the cost of getting it wrong is bad debt. No wonder teams want to automate it. It is also why a careless version is dangerous.

Where AI already runs the risk desk

AI already handles the simulation, monitoring, and bounded automation behind DeFi risk, and it is now core infrastructure. A few concrete examples, as of late 2026.

Gauntlet builds simulation-driven risk models and has acted as the economic risk brain for major lending markets, tuning parameters like collateral ratios and caps. It now runs some of the largest curated vaults too.

Chaos Labs runs AI-powered economic and security stress tests. Its risk oracles watch indicators in real time and push parameter changes within governance-approved bounds.

Aave’s renewal with the firm described more than 1,100 parameter updates.

AI-curated vaults from firms like Chaos Labs, Gauntlet, and Sentora now sit behind consumer products, rebalancing capital across Aave, Morpho, and Sky inside set risk targets.

The division of labor is simple. The machine does the speed: monitoring, simulation, anomaly detection, bounded adjustments.

The human keeps the judgment: which scenarios to run, which thresholds matter, which bounds to allow. This is real automated risk management in crypto, not a demo.

The part AI still can’t do

AI cannot be trusted to manage on-chain risk unsupervised, and 2026 showed why.

Start with the limits that never go away. It works inside human bounds, so that judgment stays human.

Models depend on data and oracles, and a stale price turns a smart system into a confident fool.

Models can be gamed by adversaries who know which inputs break them. And tail events escape training, because the crisis that matters most is usually the one the model has never seen.

The year drove this home. A string of lending blowups traced back to hardcoded or stale oracles.

In one case a collateral token was marked on-chain near $1.13 while it traded around $0.63 on the open market.

That is not a model being clever. That is automation acting fast on a wrong number.

The uncomfortable twist: automation can add risk. Fast, automated allocation paired with slow or sloppy controls, or a risk curator that itself fails, can spread a problem instead of containing it. Speed cuts both ways.

Governance-safe automation: humans still set the bounds

The norm today is not autonomous AI. It is governance-safe automation: people set the strategy and the limits, and AI acts inside them.

People decide the risk appetite and approve the guardrails. AI monitors, simulates, and makes bounded changes, often with a validation step before anything goes on-chain. And a human brake always exists.

You keep automation’s speed and keep accountability. Handing a model the power to act, with real money and real adversaries on the other side, should not be done blind.

What measured on-chain risk looks like: a Sky case study

The real answer to on-chain risk is not add AI. It is make the risk measurable. Sky is one of the clearer examples of that in stablecoins.

A quick primer, since the names trip people up. USDS is Sky’s fully backed stablecoin and the way into the system. It is the unit of account, not the yield.

sUSDS is what you hold to earn: you supply USDS, receive sUSDS, and it accrues the Sky Savings Rate over time. It is currently the largest rate-bearing stablecoin by supply.

The Sky Savings Rate is the governance-set yield behind sUSDS, funded by revenue from the Sky Agent Network, an independent set of capital allocators who put USDS to work across strategies like onchain loans, Treasury bills, and lending liquidity.

What makes it measured rather than trust me? USDS is over-collateralized and backed by diversified collateral, so the backing is not a single bet.

The yield is risk-adjusted and governance-set, not whatever the market will pay this week.

Sky is the only stablecoin protocol with credit ratings from both S&P Global and Moody’s.

Both are sub-investment grade, S&P at B- and Moody’s at B3, which is the honest part: independent agencies see real credit and concentration risk, and they publish it.

The infrastructure also carries a decade-long track record with no smart-contract exploit, and governance is building a $150 million solvency reserve.

Adoption is following. In September 2026, Galaxy Digital, a Nasdaq-listed firm, put $100 million of sUSDS on its balance sheet and approved it as institutional collateral.

And yes, this ties back to AI. Sky’s own 2026 roadmap names AI-driven automation as a growth theme, alongside Sky Agents and tokenized credit.

The through-line is the same: use quantitative, verifiable risk work, not guesswork.

If you want to go beyond the savings rate, Sky Vaults and stUSDS sit further out on the risk curve, and Sky Ecosystem Rewards layers in token incentives.

One honest caveat for savers: the Sky Savings Rate is variable. It has moved from roughly 8.75% in early 2025 to around 3.6% in late September 2026, and governance can change it. Read the live number before you decide anything.

The agent frontier, and why it’s still risky

The next step is AI agents that do not just advise but act, holding permissions and moving funds on-chain, including around tokenized real-world assets. Promising, and early.

Every limit above gets sharper when the model can pull the trigger. An agent that can act can also be manipulated or prompt-injected. A misbehaving agent causes harm directly, not as a suggestion.

Safety research keeps landing on the same answer: strict, often deterministic guardrails and clear permissions, not unsupervised control.

Full autonomy, with your money, is not here, and it is not obviously what you want.

So, can AI actually manage on-chain risk?

Yes for the parts that need speed and scale. Not yet as a manager you can leave alone.

AI already simulates, monitors, and adjusts risk faster than any human team, and it is now core plumbing for major protocols.

But it works inside human-set bounds, it can be wrong or gamed, it struggles with the crises that matter most, and letting it act unsupervised introduces fresh risk.

The grown-up version of on-chain risk is not a black box making the calls. It is measurable risk, verifiable backing, and humans who own the guardrails, with AI doing the heavy lifting in between.

AI can manage parts of on-chain risk as augmentation inside human guardrails. It is not yet a risk manager you can walk away from.

Frequently asked questions

Q. Can AI manage DeFi risk?

A. It already manages important parts, like simulation, monitoring, and bounded parameter changes, but inside human-set guardrails, not as a fully autonomous manager.

Q. What does AI actually do in DeFi risk management?

A. It stress-tests parameters, watches markets in real time, auto-adjusts limits within approved bounds, rebalances curated vaults, and flags anomalies faster than people can.

Q. What are the limits of AI risk management in DeFi?

A. It only acts within human bounds, depends on data and oracles, can be gamed, struggles with novel tail events, and over-automation can itself spread risk.

Q. How does Sky approach risk?

A. Like other protocols, Sky relies on quantitative, increasingly AI-assisted risk modeling to set the parameters behind USDS and the Sky Savings Rate, and its risk is also measured externally with credit ratings from both S&P and Moody’s. That does not make it risk-free.

• • •

On-chain risk is getting more measurable every year. That is the real story, bigger than any single model. If you want to see what governance-set, verifiable yield looks like up close, you can put your stablecoins to work on sky.money and read the live numbers on the dashboard.

AI Risk Management in DeFi: What It Can and Can’t Do was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

Every AI conversation ends the same way: close the session and the context is gone. Lighthouse Memory, the newest product from decentralized storage protocol Lighthouse, fixes that with three primitives, remember, recall and forget, persisting every memory as a verifiable blob an agent can carry across models, tools and sessions.

The Real Problem: Context Windows Aren’t Memory

Most people assume an AI “forgets” because the model isn’t smart enough to hold onto details. That’s not what’s happening. A context window is everything the model can see in the current conversation. Memory is what persists after that conversation ends. These are two different things, and most AI tooling has never actually separated them.

That’s why the same frustration repeats across every tool: start a new chat and you re-explain your project. Switch from one assistant to another and the new one is a stranger. Come back to the same thread three days later and useful context has already fragmented across sessions you can’t easily stitch back together.

The fix isn’t a bigger context window. A window that holds more still resets the moment the session ends. What’s missing is a persistence layer outside the model itself, somewhere an agent can save what matters and retrieve it later, on demand, regardless of which model or client is running at the time.

How Lighthouse Memory Works

The public surface is three verbs: remember, recall and forget. An agent connects through one of two paths, a hosted MCP server at memory-api.lighthouse.storage/mcp for any MCP-capable client (Claude Code and Claude Desktop both connect directly), or a TypeScript SDK for custom builds.

The public surface is three verbs: remember, recall and forget. An agent connects through one of two paths, a hosted MCP server at memory-api.lighthouse.storage/mcp for any MCP-capable client (Claude Code and Claude Desktop both connect directly), or a TypeScript SDK for custom builds.

Every memory that gets saved is persisted as a verifiable blob addressed by an IPFS-compatible CID, the same addressing scheme IPFS has always used. You choose the storage backend per store: Walrus, built on Sui, where memories exist as Move objects secured by erasure coding, or Filecoin, with a longer track record and continuous storage proofs (PoRep and PoSt). Retrieval works the same way regardless of which backend holds the data, recall() searches by meaning rather than exact keyword matching, and returns the matching memories with their id, CID, relevance score and tags.

Choosing a Backend: Walrus vs Filecoin

Lighthouse doesn’t force a single storage philosophy. Each memory store picks the backend that fits the workload.

As of August 2026, Lighthouse’s own pricing put Filecoin’s free tier at 5 GB against Walrus’s 100 MB, with entry-level paid plans starting around $12 a month, reflecting Walrus’s higher per-gigabyte cost for its faster retrieval. The practical split: reach for Walrus when an agent needs memories back quickly and often, and Filecoin when the priority is archival durability at the lowest possible cost per byte.

Model-Agnostic by Design

The detail that separates Lighthouse Memory from most memory products bolted onto a single assistant: it never calls a language model itself. It has no opinion about which model is doing the reasoning. An agent can run on GPT, Claude, Llama, Qwen, Mistral or DeepSeek through Ollama or vLLM, recall context from memory, and inject it directly into the prompt.

That matters practically, not just architecturally. A support agent might store “Your staging environment is at staging.acme.dev” from one conversation. Because that memory lives in Lighthouse rather than inside whichever client wrote it, a completely different model asked the same question later returns the same correct answer. Switching tools doesn’t reset what’s been learned.

Under the hood, both supported engines (a batched engine for standard use, and memwal for encrypted memory with Walrus and SEAL) implement the same interface: remember, recall, list, get, forget, status and snapshot. That consistency is what lets Lighthouse Memory function as one shared layer across every model a team happens to be using, rather than a feature locked to one vendor’s ecosystem.

What “Verifiable” Actually Proves, and What It Doesn’t

Because every memory carries an id, a CID, a relevance score and tags, an auditor can reconstruct exactly what an agent knew at the moment it made a decision, not just what it eventually did. That’s a genuine audit capability most memory layers simply don’t offer: poll status() and you can catch pending memories at session end, a keyword fallback where semantic embeddings should have matched, or a stale snapshot CID before it becomes a production problem.

Lighthouse’s own documentation is unusually precise about where that verification stops. A CID proves a memory exists and hasn’t been silently altered since it was written. It does not prove the memory was true when it was recorded. It does not prove the agent recalled the correct memory for the question it was answering. And it doesn’t establish who wrote a memory unless that’s explicitly recorded, which Lighthouse Memory does via an agent field on every entry.

That’s a narrower, more honest claim than most “verifiable AI” marketing makes, and it’s worth taking seriously for exactly that reason. Deletion semantics follow the same precision: on the Lighthouse/Filecoin path, forget() stops the storage period from renewing, the content stays technically readable until that period lapses, then is reclaimed. On Walrus, forget() removes the local copy and unpins the IPFS mirror immediately. Anyone building on this for private or regulated data should pick a backend based on that distinction, not discover it after the fact.

Getting Started

Two integration paths, depending on what’s being built.

For an MCP-capable client like Claude Code or Claude Desktop, connecting is a one-line command pointing at Lighthouse’s hosted MCP server:

claude mcp add lighthouse-memory \
  --transport http "https://memory-api.lighthouse.storage/mcp" \
  --header "Authorization: Bearer $LIGHTHOUSE_API_KEY"

Once connected, the agent gains remember, recall and forget as tools it calls on its own, deciding when something is worth saving and when a question calls for retrieving something it already knows.

For a custom agent build, the TypeScript SDK gives direct control:

import { createStorage, createEmbedder } from '@lighthouse-ai/core';
import { BatchedEngine } from '@lighthouse-ai/engine-batched';
import '@lighthouse-ai/store-lighthouse';
import '@lighthouse-ai/embed-local';

const memory = new BatchedEngine(
  createStorage(),
  { namespace: 'my-agent', embedder: await createEmbedder('local') }
);

await memory.remember('Our project uses TypeScript.', {
  tags: ['stack'],
});
await memory.flush();

const context = await memory.recall('What language do we use?');

That’s the entire surface. No vector database to provision, no schema to design before the first memory is saved.

Where This Is Headed

Lighthouse didn’t arrive at this problem from the AI side. The company spent years running decentralized storage on IPFS and Filecoin, serving a reported 2,000+ teams, with encryption, token gating and gateway infrastructure already in production before “agent memory” was a category anyone was pitching. The June 2026 Walrus integration moved from announcement to full SDK support within a single quarter, a faster shipping cadence than most infrastructure teams manage.

Lighthouse has a token generation event approaching. No official date has been confirmed as of this writing, so treat any specific timeline you see elsewhere as unverified. What’s actually worth evaluating right now is the product itself: a memory layer that’s model-agnostic, backed by a choice of two storage networks, precise about what its verifiability claims do and don’t cover, and already running in production MCP setups today. Whatever the TGE ends up looking like, that’s the part that’s live and testable right now.

Resources

• Lighthouse main website
• Lighthouse Memory
• Lighthouse Storage
• Lighthouse Blogs

Your AI Agent Has Amnesia: Inside Lighthouse Memory was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

What Is Data, and Why Does It Matter in Blockchain?

Before we talk about blockchain, DeFi, or tokenization, we need to talk about something underneath all of them:

Data.

Every transaction creates data.
Every wallet creates data.
Every blockchain records data.
But raw data is not automatically useful.
To understand what blockchain can actually do with information, we first need to understand the difference between data and information.

Data

Data is a collection of facts or information that can be processed, analyzed, and interpreted to provide meaningful insights and support decision-making.
It can exist in many forms: numbers, text, images, audio, records, and more.

What Is Data Collection?

Data collection is the process of gathering information from the world around us.
For example, a business might collect customer feedback from social media, surveys, websites, or transactions.
The goal is simple: gather raw information that can later be processed and analyzed.

Data Processing

Raw data is rarely useful in its original form.
Data processing involves cleaning, sorting, organizing, and transforming raw data into something more useful.
For example, a company might take thousands of customer responses, remove duplicate or incomplete records, organize them, and prepare them for analysis.

Data Analysis

This is where we look for patterns, trends, relationships, and useful insights.
Data analysis can involve statistics, charts, machine learning, or other analytical methods.
For example, a business could analyze its sales data to identify which products customers buy most often.

Data Management

Once data has been collected and processed, it needs to be stored, organized, protected, and managed properly.
This is where databases and information-management systems become important.
Good data management also means controlling who can access information and how that information is used.

Information

There is an important distinction between data and information.
Data is raw material. Information is data that has been processed into something meaningful.
A list of sales figures is data.
A report showing that one product is consistently outselling the others is information.
That distinction matters because information helps people make decisions.

So Why Does This Matter in Blockchain?

Because blockchain systems depend heavily on data and information.
Transactions are data.
Wallet addresses are data.
Token balances are data.
Smart-contract interactions are data.
On-chain activity produces records that can later be analyzed to understand what happened on a network.
This is one reason data is so important in Web3.
Before we talk about blockchain applications, DeFi, tokenization, or on-chain analysis, it helps to understand the basic material underneath them.

Data.

It is what gets recorded, processed, analyzed, and eventually turned into information we can use.

🤗Hello everyone web3 believers and enthusiasts, it's time to be aware of what you are dealing… was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

Artificial Intelligence and Blockchain are two of the most important tech developments of the modern era. But their routes into everyday life have been very different. AI has quickly found its way into our offices, phones, schools, search engines, creative tools, and software. Although blockchain has been around for over fifteen years, most people still find it to be far from daily life. It’s not necessarily because AI is more advanced. The thing is, AI solves a problem that people have all the time, while blockchain solves a different and more complicated problem.

AI primarily reduces the cost of intelligence and human labor. You can ask an AI to write an email, translate a document, create an image, analyze data, explain a difficult subject or help you write software. Someone can jump in and start using it and get value out of it in seconds. Blockchain is another thing. Its main advantage is that it creates trust, ownership and coordination among people or organizations that might not fully trust each other. For many blockchain applications to be useful, many participants — businesses, banks, users, regulators, merchants or institutions — need to share the same infrastructure.

There’s also a huge difference in user experience. Human language is the almost perfect interface for AI. You don’t have to learn about neural networks or machine learning to use ChatGPT. They just type what they want. The technical complexity of blockchain has historically been much more evident. The user is forced to deal with wallets, private keys, seed phrases, gas fees, network selection, bridges, addresses, and irreversible transactions. The tech requires people to learn the infrastructure before they can enjoy the product.

AI is able to also get into systems people are already using. It can be added to Microsoft Office, search engines, smartphones, Photoshop, customer-service platforms, or programming tools without changing those systems at their core. Blockchain can be even more disruptive . A blockchain-based financial system may alter the ledger’s controller, the way assets are held, the way transactions are settled, or whether an intermediary is needed at all. That means blockchain adoption is not just a question of technology. It can be a question of regulation, law, economics, institutional resistance and coordination among many organizations.

But that doesn’t mean blockchain has exactly failed. The more likely future is blockchain succeeds by becoming less visible. Most people use the Internet without knowing what TCP/IP, DNS or HTTPS are. Likewise, people might use stablecoins, tokenized assets, digital identity, decentralized settlement or blockchain-based payment systems without ever thinking about the underlying blockchain.

Perhaps the error was assuming people would desire the blockchain itself.

People don’t want a blockchain. They want faster international payments, more digital ownership, cheaper settlement, easier access to financial services, and systems that they can trust.

Why Blockchain Didn’t Become Mainstream While AI Did was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

Understanding how financial regulators are responding to AI-driven trading, investment, risk management, and decision-making.

Artificial intelligence is no longer experimental in finance. Banks, asset managers, brokers and exchanges now use AI and machine learning for trading, investment management, fraud detection, credit decisions, customer service, risk modelling and market surveillance.

Some systems help human decision-makers. Others analyse information and act at a speed no person could match.

This brings up a question every financial institution, regulator, and customer should be asking:

When an algorithm makes the decision, who is legally responsible for it?

The question is more difficult than it seems.

A decision that is made, shaped, or carried out by an algorithm does not automatically change the legal duties attached to it, yet it can blur who is expected to answer for the outcome.

Is it the institution that deployed the system, the team that approved it, or the vendor that built it?

Across jurisdictions, the approach differs. Many regulators have started from existing legal principles: applying financial regulation regardless of whether financial activity becomes automated. Some are going further by introducing AI-specific legislation, while others are relying on regulatory guidance to set out regulatory obligations.

To see how this works in practice, let’s examine what regulators expect from firms, how different jurisdictions are responding, and where the greatest risks sit.

Applying Existing Rules To New Technology

Financial regulation has always focused on activities and outcomes. A securities regulator cares about market integrity whether a trade was placed by hand or an algorithm.

An investment adviser has the same duties to clients whether or not AI helped produce the advice. A bank must manage credit and operational risk whether the assessment comes from a person or a model.

AI changes how the work is done. It does not change who answers for it.

That is why regulators keep asking the same practical questions about any AI system.

Here is each one, with what a regulator expects to hear:

Who approved the system?

A named, senior person or committee should have signed it off, usually after review by risk, compliance, and legal teams.

“The vendor said it was fine” is not an approval.

What data does it use?

The firm should have a documented list of every data source, where it came from, whether the firm is legally allowed to use it for this purpose, and how good its quality is.

What decisions can it make?

The firm should define the system’s limits in writing, such as the maximum trade size, the types of customers it can assess, or whether it can only recommend rather than act.

How is it tested?

Before launch, models should be validated against historical data, stress-tested in unusual market conditions and checked for bias. After launch, testing should continue, because models can drift as markets change.

Who monitors its performance?

The team that owns the system should track it daily, and an independent risk or audit function should review it separately.

Can its output be challenged?

Customers should have a route to ask for a human review of significant decisions, such as a declined loan, and staff should be able to override the system.

What happens when the model produces an incorrect or discriminatory result?

There should be an incident process: pause or limit the system, find the cause, fix it, correct the affected decisions, compensate customers where needed and report to the regulator when the rules require it.

Can the institution reconstruct why a decision was made?

It should be able to, through records of inputs, outputs and model versions. If a firm cannot explain a decision to a regulator or a customer, it has a governance problem.

Because of these questions, AI governance is a financial regulation issue, not just a technology issue.

The Risk Depends on What the AI Does

Not every AI use carries the same risk.

The right starting point is the function, not the technology.

• Summarising internal documents: the main concerns are accuracy and confidentiality.
• Investment advice: raises questions about suitability, conflicts of interest and disclosure.
• Credit scoring: raises concerns about discrimination, data protection and explaining why someone was declined.
• Market surveillance: risks include false alarms and missed suspicious activity.
• Automated trading: raises market-abuse, operational and resilience risks.

The focus of this inquiry is on the role of AI: What does the AI do, what decisions does it influence, and what happens as a result?

How Different Jurisdictions Are Responding

Australia

The Australian Securities and Investments Commission (ASIC) has urged firms to check that their use of AI is consistent with existing obligations, including directors’ duties, risk management and consumer protection.

The practical message is that bringing in an AI system does not shift responsibility from the firm to the technology.

Canada

Canadian securities regulators have focused on transparency, accountability and risk management when capital-market participants use AI, particularly where decisions affect investors or clients.

European Union

The EU shows the layered approach most clearly. Financial firms already follow detailed rules on market conduct, risk management, outsourcing, operational resilience and consumer protection.

These now sit alongside the EU’s AI Act, which sets its own requirements depending on how risky an AI use is. Credit scoring of individuals, for example, is treated as higher risk.

This means a firm can comply with the financial rules for an activity and still owe extra obligations because AI is performing it. Firms need to check both.

China

China has taken a more prescriptive approach, with regulatory attention on algorithm governance, data management, risk controls and financial stability. Its concern goes beyond individual customers. If many institutions use similar models, data or technology providers, one fault could spread across the system. That turns AI into a question of financial stability as well as compliance.

United States

The US has no single AI statute for financial services. Instead, AI-enabled activity falls under existing agencies and laws, including the Securities and Exchange Commission, the Commodity Futures Trading Commission, and banking and consumer-protection regulators. If an AI system facilitates misleading conduct, market manipulation, unsuitable advice or weak risk controls, the use of an algorithm is no defence. The technology may be new. The legal obligation is not.

What If the AI Comes From Someone Else?

Many financial firms buy AI from cloud providers, software companies and data vendors instead of building it.

So who is responsible when a vendor’s model fails?

The financial firm is.

Outsourcing the technology does not outsource the regulatory responsibility. The vendor may owe the firm compensation under their contract, but the regulator will still look to the firm.

To manage this, firms should do the following:

• Ask how the vendor trained and tested the model
• Find out what data the model can access.
• Agree how performance is monitored and how model changes are announced in advance.
• Secure contractual rights to audit the system and to receive the information needed to explain its decisions.
• Keep a plan for switching vendors or turning the system off if something goes wrong.

The more central the AI is to a regulated activity, the less it can be treated as an ordinary software subscription.

Data Is Part of the Problem

AI models may process transaction histories, credit records, customer profiles, communications and behavioural data.

A model can be accurate and still break the law if it uses data without a proper basis, uses it for a different purpose than the one it was collected for, collects more than it needs, stores it insecurely or keeps it too long.

This matters most when AI makes decisions about people. Accuracy is only part of the test. The data and the decision process must also be lawful and well governed, and people affected should be able to exercise their rights, such as asking for an explanation or a correction.

Human Oversight Still Matters

Human oversight does not mean a person must approve every AI decision. The right level depends on the system and the cost of failure. In practice there are three common models:

• Human in the loop: a person approves each decision. Suitable for high-stakes calls, such as large credit decisions.
• Human on the loop: the system acts on its own, but a person monitors it and can step in. Common in trading, with limits and alerts.
• Human in command: people set the rules, test the system, and can switch it off at any time. This should apply to every system.

Whatever the model, firms need real ways to monitor, test, challenge and override their systems. A model that can execute trades or approve transactions without effective controls carries far more risk than one that simply assists an employee.

The Market Integrity Problem

AI can analyse markets, build strategies and trade far faster than any human. That can improve efficiency, but it can also magnify misconduct and operational failures at speed.

A strategy does not become legal because an AI designed it, and a system that produces misleading information or enables manipulation cannot use automation as a defence. This is why regulators continue to focus on algorithmic trading controls, market surveillance and market-abuse rules. Technology changes the speed and scale of conduct, not the legal standard.

Ten Questions Every Financial Institution Should Ask

1. What financial activity does the AI support? Name the exact activity, such as credit approval or order execution, because the rules follow the activity.
2. What decisions can the system make or influence? Write down its authority and limits, and say whether it recommends or acts.

3. What regulatory obligations apply to that activity? Map the financial rules, data protection law and any AI-specific law that applies in each country of operation.

4. What data does the system process? Keep an inventory covering source, legal basis, quality and sensitivity.

5. How was the model tested and validated? Use independent validation before launch, plus stress and bias testing, and re-test regularly.

6. Who is responsible for monitoring it? Assign a named owner, plus an independent second line of review.

7. Can the institution explain or reconstruct important decisions? Keep logs of inputs, outputs and model versions so any decision can be traced.

8. What happens when the model produces an incorrect result? Have an incident plan covering suspension, root-cause analysis, customer remediation and regulator notification.

9. How are third-party AI providers governed?Use due diligence, audit rights, change notifications and an exit plan.

10. Can the system be stopped or overridden when necessary? Build and test a kill switch, and make sure someone is authorised to use it.

These questions are quickly becoming part of ordinary financial governance.

Where Regulation Is Heading

AI regulation in finance will not follow one universal model. Some countries will lean on existing financial rules, some will pass dedicated AI laws, and many will do both. However, the principle is increasingly clear: AI does not create a regulatory vacuum.

A financial institution remains responsible for what it does, even when technology performs most of the work. The question that matters is how the system is used, what decisions it influences, what risks it creates and who is accountable when something goes wrong.

That is where AI governance becomes financial regulation.

If you enjoy analytical commentary on digital asset regulation, crypto markets, and emerging financial technologies, consider subscribing to my newsletter where I share additional research, commentary, and industry insights.

https://samuel-ayodeji.kit.com/profile

Also, if your company, startup, or publication needs clear, well-researched content on blockchain, digital assets, fintech, or emerging technology law, my inbox is always open.

AI in Financial Markets: Who Is Responsible When the Algorithm Gets It Wrong? was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

Index funds made investing simple. Grid bots made it automated. Agentic AI is trying to make it adaptive. Here’s what actually separates them.

Your index fund doesn’t know what the market is doing. Neither does your grid bot.

One buys everything and waits. The other places orders on a fixed ladder and hopes price keeps bouncing inside it. Both are automated, and both are blind to context.

That’s the uncomfortable truth behind the most common question passive investors are asking right now: is buy-and-hold indexing still the best risk-adjusted strategy, or is there something smarter for hands-off investors?

Over the past few years, two “smarter” options have competed for your attention: rule-based grid trading bots, and a newer category called agentic AI. They’re often lumped together as “automated trading.” Structurally, they have almost nothing in common.

This article breaks down exactly how they differ, why it matters for your portfolio, and why platforms like Hyperlyx AI are built on the agentic model rather than the rigid bot model.

Grid bots execute fixed rules and can’t adapt when markets change regime. Agentic AI systems observe conditions, reason about risk, and adjust their own behavior within guardrails. For hands-off investors, that difference is the gap between a tool that works only in the market it was built for and one designed to handle the market you actually get.

What Is a Grid Bot? (And Why Did So Many Investors Adopt Them?)

A grid trading bot is an automated program that places buy and sell orders at preset price intervals above and below a set price, forming a “grid.” When price falls to a lower rung, it buys. When price rises to a higher rung, it sells. Each completed cycle captures a small profit.

The appeal is obvious:

• It’s automated: Set it and walk away.
• It’s intuitive: “Buy low, sell high” repeated mechanically.
• It thrives in sideways markets: Choppy, range-bound price action is a grid bot’s ideal habitat.

For a while, that made grid bots feel like a cheat code. In a range, every wiggle becomes income.

But the strengths of a grid bot are inseparable from its weaknesses. It performs well in exactly one kind of market, and markets rarely stay in one kind of market.

The Structural Flaws of Grid Bots

The word that matters here is structural. These aren’t bugs a software update fixes. They come from how grid bots are designed.

1. They Are Rule-Bound, Not Context-Aware

A grid bot follows the parameters you gave it at setup: range, grid spacing, order size. It doesn’t ask why price is moving. A slow drift down and a panic-driven collapse look identical to it, and it responds identically to both.

2. They Break on Trends

If price breaks out above your range, a grid bot sells everything on the way up and then sits in cash while the trend runs without it. If price breaks below your range, the bot keeps buying the way down, accumulating a growing position in a falling asset. Trending markets, which are where much of the long-term return in investing comes from, are the grid bot’s blind spot.

3. They Require Constant Human Babysitting

The irony of the grid bot is that it markets itself as hands-off but demands supervision. Someone has to redraw the range, resize the grid, and decide when to shut it down. That someone is you, which defeats the entire purpose for a passive investor.

4. They Have No Real Risk Layer

Most grid bots have, at best, a blunt stop-loss. They don’t reason about portfolio-level exposure, correlation, volatility regimes, or drawdown limits. Risk management is something you bolt on manually.

5. They Are Backward-Looking by Design

Grid parameters are typically tuned to recent price behavior. That’s a bet that the near future will look like the recent past. Every experienced investor knows how that bet ends when conditions shift.

None of this makes grid bots useless. It makes them narrow tools being sold as general solutions.

What Is Agentic AI?

Agentic AI refers to artificial intelligence systems that can pursue a goal autonomously: perceiving their environment, making decisions, taking actions, and adjusting based on results, all without step-by-step human instruction.

The key word is agency. A traditional bot follows a script. An agentic system works toward an objective.

In an investing context, an agentic AI system typically does four things a grid bot cannot:

1. Observes: ingesting market data, volatility conditions, and shifting patterns rather than a single price range.
2. Reasons: weighing whether current conditions favor more exposure, less exposure, or no action.
3. Acts: executing decisions within defined risk boundaries.
4. Adapts: updating its behavior as regimes change instead of waiting for a human to reconfigure it.

That loop of observe, reason, act, adapt is the structural heart of the difference.

Agentic AI vs. Grid Bots: The Core Structural Differences

Here’s the side-by-side view.

The through-line is adaptability. Grid bots are engineered around one market assumption. Agentic systems are engineered around the assumption that the market will keep changing.

The Real Question: Is Passive Indexing Still Optimal?

Before pitching anything, it’s worth being fair to the strategy most readers are currently using.

Is standard index investing still a good strategy? For most long-term investors, yes. Broad-market index funds offer low fees, diversification, and tax efficiency, and decades of research (including the SPIVA reports) show that a majority of actively managed funds fail to beat their benchmarks over long horizons. Anyone telling you indexing is “broken” is overselling.

But passive investors ask a more precise question: is buy-and-hold optimal on a risk-adjusted basis? That’s a different question, and it has a more interesting answer.

The Hidden Cost of Pure Buy-and-Hold

Indexing gives you the market’s return, and also the market’s drawdowns. Broad indices have historically dropped 30 to 50 percent in severe bear markets, and recovery can take years. Two things follow:

• Sequence risk: If a major drawdown lands right before or during your withdrawal years, the damage can be permanent.
• Behavioral risk: Many “passive” investors aren’t passive when their portfolio is down 35 percent. They sell. Buy-and-hold only works if you actually hold.

So the real weakness of passive indexing isn’t returns. It’s that it has no mechanism to respond to risk. It’s designed to ignore conditions entirely, which is a feature until it isn’t.

This is where the industry sees an opening: not to replace long-term investing, but to add an adaptive risk layer on top of it.

Where Hyperlyx Fits: The Next-Generation Upgrade

This is the gap Hyperlyx AI is built around.

Grid bots are automation without intelligence. Index funds are discipline without adaptability. Hyperlyx AI positions itself as the third path: an agentic AI approach designed for hands-off investors who want automation that thinks.

Instead of asking you to define a price range and hope for the best, an agentic architecture is built to:

• Read changing market conditions rather than a static price band
• Adjust exposure dynamically instead of buying and selling on a fixed ladder
• Prioritize risk management as a core function, not an afterthought
• Reduce the maintenance burden that makes traditional bots anything but hands-off

The pitch isn’t “beat the market every year.” No honest platform should promise that. The pitch is that a system capable of adapting is structurally better suited to the real world than one that can only repeat.

If grid bots are a vending machine, agentic AI is closer to a portfolio analyst who never sleeps, never panics, and works within boundaries you set.

Want to see the full framework? Read the Full Agentic AI Blueprint PDF to see how the architecture works under the hood.

5 Questions to Ask Before Trusting Any Automated Investing System

Whether you’re evaluating Hyperlyx, another agentic platform, or a bot, use this checklist. It works for any automated strategy.

1. What happens when the market regime changes?

If the answer is “you’ll need to reconfigure it,” it isn’t hands-off.

2. Where does the risk management live?

It should be inside the system’s logic, not a manual setting you have to remember.

3. How does it behave in a trending market?

Ask specifically about strong uptrends and sustained downtrends, not just sideways action.

4. Is performance shown across multiple conditions?

Beware of backtests cherry-picked from one favorable period. A strategy that only looks good in one regime is a grid bot in disguise.

5. Who’s accountable for the guardrails?

You should be able to set and understand your risk limits in plain language.

A system that answers all five well is worth serious attention. A system that dodges them is selling you a story.

Common Questions About Agentic AI and Grid Bots

What is the main difference between agentic AI and a grid bot?

A grid bot executes fixed, pre-set rules within a defined price range. Agentic AI pursues a goal by observing conditions, reasoning about risk, and adapting its actions over time.

Are grid bots a good strategy?

They can perform well in stable, range-bound markets. They tend to struggle in strong trends and require frequent manual adjustment, which limits how hands-off they really are.

Is passive index investing still the best strategy?

For most long-term investors, low-cost indexing remains a strong, evidence-backed foundation. Its main limitation is that it has no built-in response to drawdowns or changing risk conditions.

Can agentic AI replace index funds?

Not necessarily. Many investors treat agentic strategies as a complement, adding an adaptive risk-management layer to a long-term core strategy rather than abandoning it.

Is agentic AI investing risky?

All investing carries risk, including loss of principal. Agentic systems aim to manage risk more actively, but no system eliminates it or guarantees returns.

What does “hands-off investing” actually mean?

It means minimal ongoing involvement from you. True hands-off investing shouldn’t require you to constantly retune settings, redraw ranges, or monitor positions daily.

The Bottom Line: Automation Isn’t the Same as Intelligence

The confusion driving this whole conversation is simple. People hear “automated” and assume “smart.” They’re not the same thing.

• Index funds automate discipline.
• Grid bots automate repetition.
• Agentic AI aims to automate judgment, within the limits you define.

For passive investors asking whether the old playbook still holds up, the honest answer isn’t “throw it out.” It’s that the old playbook has a gap, the gap is risk that changes over time, and only an adaptive system is built to address it.

That’s the structural difference, and it’s why the conversation is moving from rigid bots to agentic AI.

Ready to go deeper? The full breakdown of how the agentic architecture works, including the risk framework and how it compares to traditional approaches, is in the free guide.

Read the Full Agentic AI Blueprint PDF

If this article changed how you think about hands-off investing, clap so Medium shows this to more readers and repost or share it with the friend, colleague, or family member who still thinks “passive” means “no thinking required.” They’re exactly who this was written for.

Follow for more on agentic AI, risk-adjusted investing, and the tools reshaping how ordinary people build wealth.

Agentic AI vs. Grid Bots: Structural Differences Every Hands-Off Investor Must Know was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.