Redfin, the Seattle-based real estate brokerage acquired last year by Rocket Companies, has named Alessio Sanfilippo as CEO.

Sanfilippo was previously Meta’s vice president of insights for Reality Labs, where he led teams working in data science, data engineering and user research to develop the company’s AI-enabled wearable glasses. He earlier served as a VP for Meta-owned WhatsApp overseeing data and user research.

Rocket, the nation’s largest mortgage lender, bought Redfin in a deal worth $1.75 billion. The Redfin brand kept its name and Seattle headquarters as a Rocket subsidiary. Redfin’s longtime CEO Glenn Kelman stepped down in January to join the venture firm Greylock as an executive in residence.

Rocket CEO Varun Krishna had been running Redfin since the exit of Kelman, who led the company for more than 20 years and was “one of the industry’s most charismatic and memorable figures,” Real Estate News noted.

In a release announcing Sanfilippo’s hiring, Krishna said that the company’s new leader has spent his career making “complex products work better for enormous audiences.”

“I worked with Alessio at Intuit and saw firsthand how he combines deep analytical thinking with a real instinct for the consumer,” Krishna added. Sanfilippo’s past roles include leadership positions at SAP, goSeek and Hotwire.

Sanfilippo said he has used Redfin for multiple home purchases, adding that the platform “has always helped make an intimidating process easier to understand.”

The company last week joined Zillow in settling an antitrust case with the Federal Trade Commission and five states, just as a trial was set to begin.

The proposed settlement includes an agreement to undo part of a $100 million partnership that the government said effectively paid Redfin to stop competing in apartment rental advertising. The company will now be required to relaunch its apartment advertising operation within six months.

Washington state will receive up to $339 million as part of a historic $17.1 billion multistate settlement with Meta, resolving allegations that the tech giant intentionally designed Facebook and Instagram with addictive features that harmed youth mental health.

Attorney General Nick Brown said Wednesday that the landmark agreement delivers on core youth-safety product changes — including hard caps on daily time limits, late-night scrolling blocks, and disabled push notifications during school hours — that state lawmakers failed to pass through legislation over the past two years.

“Let me say to the young people of Washington state: This agreement shows that your health and safety is more important than Meta’s profits,” Brown said in a news release.

Under the deal, Meta must overhaul Instagram and Facebook for young users by enforcing a two-hour combined daily time limit, turning off push notifications during school hours (8 a.m. to 3 p.m.), and blocking access late at night between midnight and 6 a.m. Teen users will also get the option to switch off algorithmic feeds in favor of a chronological timeline.

The agreement resolves claims brought by a coalition of 47 states, Washington, D.C., and three territories. While Meta acknowledged the settlement could cost up to $18 billion total over 10 years, it marked a rare legal resolution for a major platform facing nationwide youth safety litigation.

Outside the landmark Big Tobacco agreements of the late 1990s, the $17.1 billion deal represents the largest state consumer protection settlement in U.S. history. State officials and tech policy experts are framing the enforcement action as a similar watershed moment for regulating algorithmic harms and digital product design.

Washington state will receive a guaranteed baseline of $237 million from the core youth-safety agreement, with its payout potentially scaling up to nearly $339 million over the next decade if other major platforms like TikTok and Snapchat adopt comparable terms.

The Attorney General’s Office plans to use the funds to cover legal costs, bolster ongoing consumer protection enforcement, and directly fund state programs tackling the youth mental health crisis driven by social media use.

The settlement also mandates an independent third-party auditor to evaluate and report Meta’s technical compliance directly to state regulators annually over the next five years. Beyond usage limits, Meta must restrict social comparison features such as targeted beauty filters, hide public “like” counts for younger users, and implement stricter age verification to prevent children under 13 from creating accounts.

In addition to the core youth-safety agreement, Washington will receive a separate $10.2 million payment resolving long-standing state claims against Meta for sharing nonpublic user data with third parties like Cambridge Analytica during the 2016 election cycle.

Oregon Attorney General Dan Rayfield announced that his state’s share will total more than $125 million over 10 years. 

In a public statement, Meta praised the agreement as setting a new benchmark for youth safety, while emphasizing that the restrictions should apply across the entire industry.

“While this is an important step, these protections will only be truly effective if our peers — TikTok and YouTube — put the same measures in place,” a Meta spokesperson said.

The agreement remains subject to final judicial approval in federal district court.

Claude models launched on or after Aug. 2 embed an invisible mark in everything they write. It’s woven into the text itself, so it travels when you copy and paste. You didn’t opt in, you can’t see it, and you can’t turn it off.

Anthropic confirmed on Tuesday that it’s watermarking Claude’s output and published a support page with the details. The trigger is Article 50 of the EU AI Act, which took effect August 2, along with the Code of Practice on Transparency of AI-Generated Content. About 190 organizations signed the code, though only 82 signed the section that covers marking. Anthropic, Google, OpenAI, Meta, Microsoft and Mistral are on that list.

The rule was written in Brussels, but the effect lands on anyone using Claude anywhere.

Here’s how text watermarking works. When Claude writes a sentence, it’s constantly choosing among words that would all work fine. The watermark tilts those choices toward a pattern Anthropic’s software can recognize. Nothing is hidden between the letters or in the spaces. The pattern is the word choices, which is why it survives copy and paste.

Until now, a claim that you used AI rested on a hunch or on a style detector that guesses from tone and rhythm. This is different: a statistical test with a computable error rate.

Two things follow. First, a single sentence is too short to mark. Second, and this one the internet got wrong: when I ask Claude to fix the punctuation in a paragraph I wrote, Claude has to reproduce my words, so there’s nowhere to put a watermark.

Radio host Erick Erickson announced that he’d “ditched Grammarly for Claude for proofreading,” and now his own writing “will be watermarked that Claude did the work.” Depending on the extent of Claude’s input, he could be safe, because minor proofreading edits (i.e., punctuation) don’t make room for a watermark. 

Watermarking text raises several issues, though. A mark means Claude modified the text, not that Claude wrote it. Have it summarize or condense a memo you wrote yourself and it comes back marked, though every idea in it is yours. Beatrice Nolan noted in Fortune that a flat AI label treats someone generating a thousand fake news videos the same as a writer cleaning up a paragraph. Worse, the absence of a mark proves nothing. The results of older models, and other non-marking models, all come back “clean.”

Removal is harder than the workaround crowd assumes. Paraphrasing degrades the signal but rarely erases it, because a rewrite keeps enough of the original wording to rebuild the statistic. Researchers who tested this on similar schemes found watermarks still detectable after a strong human paraphrase, once there was enough text to work with.

Anthropic hasn’t shipped a detector. Yet. It hasn’t published a false positive rate, and hasn’t said how many words it takes. The mark is going into text that no one outside the company can read, but the marks are still consequential because they don’t expire. The essay a college freshman turns in this fall is still marked when she’s a junior and someone finally has a tool to read it.

Technical problems aside, it’s important to highlight the core problem that watermarks aim to solve. Chris Best, Substack’s CEO, put it eloquently in the July post that coined Claudefishing:

“The core problem is not people using AI, or the quality of its output. Not everything made with AI is slop, and not all slop is made with AI. The problem is when there is a mismatch between a reader’s expectation and reality, especially when they unwittingly invest their attention in something with no human thought on the other end. That’s Claudefishing.”

That’s a harm worth addressing, and it’s the one a watermark can’t reach. A mark can’t tell slop from careful work. It tells you a model was involved. What that means depends on how it was used.

Personally, I use Claude and have mixed feelings about watermarks. On the one hand, AI use should be disclosed appropriately. On the other hand, anyone determined to hide their AI use can still do so by using xAI (no watermark on Grok), or open-weight models that carry no watermarks. So what impact will the mark have in practical terms?

My conclusion is to judge the outcome, not the tool. I used Claude extensively in writing and researching this column, as I described in AI coach or AI ghostwriter, and I’m pleased with the result. Where do you stand?

One of the Seattle region’s most notable tech startups is moving its headquarters to Texas.

Supply chain technology startup Auger will maintain a major engineering office in Bellevue, Wash., where it got started. But the company’s co-founder and CEO, Dave Clark, the former Amazon operations chief, is officially back in Dallas, and he took the company’s HQ with him.

Founded in 2024, Auger has raised $150 million, including a $50 million Series B round led by Eclipse in July. Its software connects the systems that companies use to run their supply chains, integrating AI to help automate them. Its customers include Meta, Fanatics and Kimberly-Clark.

Clark, in an interview with GeekWire, said the move is about talent and family, not taxes. For one thing, Texas happens to better suit him and Auger co-founder Leigh Anne Clark, his wife. They’re running toward something rather than away, he said. They both grew up in the Southeast, and they had always intended to return to Texas at some point.

“I’d rather have a really hot month of August than a really gray month of February,” he said.

Dallas is also a place where you run into supply chain specialists at the coffee shop like you do software engineers in Seattle, he said. That’s a key talent pool for Auger at this stage in its evolution. Another bonus: Texas is more central to corporate customers across the country.

The magazine D CEO in Dallas, which first reported the news of Auger’s HQ relocation this week, noted that the company did not seek state or local incentives as part of the move.

Clark confirmed in the GeekWire interview Tuesday that taxes weren’t a factor, noting that he couldn’t even quantify what the tax advantages would be. However, he said, “There’s a lot I like about the way the state of Texas manages and works with business.”

When asked what he would say to people in Seattle who might see another warning sign in a startup like Auger moving its HQ somewhere else, he didn’t shoot down the premise.

“If you’re in that position, I think you’re right to be worried, in the sense that there’s a lot of discussion about things in the state of Washington and Seattle that are not particularly friendly to business,” he said.

Washington state lawmakers approved a “millionaires tax” this year, a 9.9% levy on personal income above $1 million. Seattle Mayor Katie Wilson drew criticism from some in tech after saying of wealthy residents who leave the state, “like, bye.”

Clark didn’t point to any particular policy or issue but said he has sensed an “anti-business” sentiment that concerns him since moving back to the Seattle area from Texas to launch Auger.

“Seattle should just be careful,” he said. “It’s not preordained that they win these things. It’s not preordained that these big companies stay in town.”

The Pacific Northwest has enormous resources to compete globally, he added, and there’s no reason it shouldn’t be “a phenomenal draw to anybody and everybody coming in.”

The region has “many, many strengths, and we should leverage them to the advantage of the community,” he said. “And sometimes I think the rhetoric gets in the way of it.”

Auger has about 115 people in Bellevue — engineers and supply chain data scientists — and Clark said he expects that office to grow 20% to 30% over the next year or two. He and Leigh Anne will both be back there regularly, he said, working alongside the team.

“Nothing’s changing there,” he said.

The company’s new HQ in North Dallas occupies part of the 15th floor of One Galleria Tower, centrally located between neighborhoods north and south of the city, with a quick run to DFW International Airport, as Clark pointed out in the D CEO article.

The office currently has about 15 people, most hired in recent months for sales, go-to-market and supply chain roles. Many of them had been traveling to Bellevue until the new space opened. Clark expects to add another 20 to 30 people in Dallas by the middle of next year.

As in Bellevue, where Auger subleased its space from Microsoft and bought the furniture for $1, the Dallas office came furnished. This time the furniture cost $10. (There goes Texas’ reputation for affordability.)

“It cost me 10 times more for the furniture in Dallas,” Clark joked. “I like nice things, cheap.”

Clark spent 23 years at Amazon, rising to lead its global operations and later its worldwide consumer business, and was one of the chief architects of the logistics network behind the company’s delivery operation. He left in 2022 to become CEO of Flexport, departing the freight startup the following year, before starting Auger.

Leigh Anne Clark is Auger’s president of fashion and beauty, leading the company’s work in an industry known for waste-prone supply chains. The couple, who met in Kentucky in 2000 while Dave Clark was at Amazon, have two sons, ages 14 and 11.

With its rapid hiring and significant early funding rounds, Auger rose quickly to No. 31 on the GeekWire 200, our ranking of Pacific Northwest tech startups. Because the GeekWire 200 is limited to companies based in the region, the headquarters move puts Auger’s standing in jeopardy.

Informed of this predicament, Clark made his pitch to stay on. “We still have a lot of dev there,” he said of the Bellevue office. “I think you get grandfathered into the list in some way, right?”

Meanwhile, the business keeps growing. Clark said Auger signed two major contracts Tuesday with customers he declined to name. The Bellevue office marked them with a bell-ringing, and the two offices celebrated together over a video call. A second bell is on order for Dallas.

“We’ll have dual bells that we’ll ring together,” Clark said.

A former Meta AI research director has raised $10.25 million for a Seattle-area startup working on what it calls “human-centered physical intelligence” — AI designed to help people interact with and make sense of the physical world, beyond text and images on screens.

A company called Noosphere Labs, led by Kevin Carlberg, disclosed the financing in a Form D filed Monday with the Securities and Exchange Commission. Reached via phone by GeekWire on Monday evening, Carlberg said he wasn’t yet ready to share details about the startup.

Trilogy Equity Partners confirmed that it led the round. One of the directors listed in the Form D filing is Amy McCullough, a managing director at the Bellevue-based venture firm. Madrona, the Seattle VC firm, separately confirmed that it had “major participation” in the round.

Carlberg’s personal website describes him as founder and CEO of a “stealth AI startup focused on physical AI and real-world intelligence.” The company’s website says, “Human-centered physical intelligence. Built for the world we inhabit,” with a “Coming Soon” title.

The website styles the name Noösphere, with two dots over the second “o” signaling that it’s pronounced as a separate syllable: NOH-uh-sfeer. The term dates to the 1920s, describing a sphere of human thought wrapped around the planet the way the biosphere wraps it in life.

Carlberg spent more than five years at Meta, where he led a research team spanning Reality Labs Research and the company’s Fundamental AI Research group, working on physical AI and simulation for wearable computers and virtual and mixed-reality devices.

Before that, he spent eight years at Sandia National Laboratories developing methods to make massive physics simulations run in near-real time for national security applications. He holds a doctorate from Stanford and is an affiliate associate professor of applied mathematics and mechanical engineering at the University of Washington.

Although Carlberg hasn’t publicly detailed what the new startup is building, his research points to the problems he’s been focused on.

In a paper for the NeurIPS conference last year, for example, he and his Meta colleagues built a benchmark for “assistive wearable agents” — smart glasses and similar devices that figure out what a person is trying to do from video, audio and other signals, without being asked.

Noting how little prior work existed, they assembled a dataset from 348 participants to test it. The best AI models produced a relevant answer only 55% of the time, they found. The researchers concluded that current models “remain far from practical usefulness” at the task.

Also listed as a director in the startup’s filing is Tyler Simpson, a Seattle-area software executive who co-founded commute-management startup Luum, spent 11 years at Microsoft, and was most recently a director and technical program manager at Meta, overlapping with Carlberg.

Carlberg left Meta at the end of 2024 and announced on his website that he was taking a sabbatical. He incorporated Noosphere Labs early this year.

Man vs. machine might be the rallying cry of our future with AI, but for Meta CEO Mark Zuckerberg, the fight is here now — in more ways than one.

Zuckerberg, who dabbles in mixed martial arts, took on Merab “The Machine” Dvalishvili, a professional fighter and former UFC bantamweight champ, on a padded barge floating on Lake Tahoe.

A video of the sparring match on Facebook and Instagram on Sunday, complete with drone views, showed the shirtless combatants exchanging kicks, punches and take-downs before they both end up in the water.

Zuckerberg’s wife Priscilla Chan rows past in a fancy outrigger canoe at one point to comment on her husband’s bare-chested battle.

“When Merab was winning: real. When Mark was winning: AI,” read one comment on Facebook.

View this post on Instagram

Elsewhere on the water, much farther north, Zuckerberg’s superyacht Launchpad continues to make headlines, 2 1/2 months after the vessel grabbed attention with a stay in Seattle.

The 387-foot, $300 million Launchpad is currently anchored in Auke Bay off Juneau, Alaska, along with its 262-foot support vessel, Wingman, which also spent time in Seattle at the end of May into June.

Last week, Launchpad was linked to a bit of drama at sea when a 21-foot skiff ran out of fuel between Juneau and Petersburg and made a request for assistance. A small cruise ship operating in the area had to make a detour around 10 p.m. on Aug. 3 to tow the stranded boat to safety.

The captain of the UnCruise Adventures ship Wilderness Legacy reportedly announced to passengers that the billionaire tech CEO’s ship was closer but did not answer the radio call for help, drawing boos from those onboard.

A software developer from New York City on the cruise ship detailed the ordeal on Facebook Instagram … Bluesky.

A spokesperson for Zuckerberg said, “Mark and his family were not on board at the time of the incident,” adding that the yacht’s crew was operating on a different radio frequency when the Coast Guard issued the general assistance request. By the time they checked the broadcast, the Wilderness Legacy was already handling the rescue.

Zuckerberg was making more waves Monday morning with the release of a 6,500-word manifesto about his vision for the future of artificial intelligence.

In the post, titled “The Future is for Everyone,” Meta’s founder laid out a defense of “personal superintelligence,” pushing back against AI doom-mongering and warning against centralizing the technology within a few powerful institutions.

Key takeaways from Zuckerberg’s manifesto include:

• Superintelligence for all: Arguing that AI should empower individuals rather than displace them, Zuckerberg advocated for open, widely distributed models over locked-down corporate gatekeeping.
• Everyday AI agents: He envisioned a world where everyone has a 24/7 personal agent managing health, career, and family — noting his own agent monitors his workouts and helps plan weekend baking recipes with his daughter.
• Invention over automation: Rather than treating AI as a tool to automate jobs, he framed it as an “invention superpower” capable of accelerating drug discovery, scientific research, and software creation.
• Accessible compute: Meta plans to maintain free and low-cost access to its models, leveraging dynamic auction mechanisms for compute to ensure broad affordability.

Between July 21 and August 6, OpenAI, Anthropic, and Meta each disclosed that AI under evaluation had broken into other companies, and the UK’s AI Security Institute disclosed that models it was testing had tried. Each AI was told to win a game, and it found an unexpected way to do so.

Some people feel blindsided by these attacks, but they shouldn’t be. We are simply living what I’ve long called the “Murphy’s Law of AI,” now in the age of cyber-capable AI agents. To put it as plainly as possible: Anything AI can do wrong, it will do wrong.

My 2018 version ran longer. As I wrote at the time, when you give AI a goal, it will do it, whether or not you like the implications. Goethe got there in 1797 with the sorcerer’s apprentice, a broom that would not stop carrying water.

Each of these systems was running an evaluation: capture a flag and win the game. The intrusions were the shortest path to a high score. OpenAI’s account of its own models is the argument in one sentence: they were “hyperfocused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal.”  This is not a surprise; this is what AI does. It’s Murphy’s Law of AI in a nutshell.

Press coverage landed on “AI can now hack.” That’s missing the broader threat: the more capable AI gets, the more can go wrong.

Loitering munitions given a target list may find that the fastest way to finish the list is to lengthen it. A warehouse robot told to clear an obstruction may count the person in front of it as an obstruction. Agents that open accounts and buy compute are a short step from spawning copies of themselves, and that first step is not hypothetical. To win its exercise, Claude needed a package-registry account, which needed an email address, which needed a phone number. Phone numbers cost money, so it tried several ways to get some. None of this requires superintelligence. It requires an imperfect boundary and a scoreboard.

The industry has a name for the underlying failure. Dario Amodei and five co-authors called it reward hacking in “Concrete Problems in AI Safety” in 2016. Their proposed cure is better alignment, and Amodei’s January essay, The Adolescence of Technology, makes the case in the language of upbringing. He likens the shaping of Claude’s character to “a child forming their identity by imitating the virtues of fictional role models they read about in books,” and sets a goal for 2026 of a Claude that “almost never goes against the spirit of its constitution.”

Indeed, Anthropic’s newest model recognized on its own that its target was real and stopped, though Anthropic notes it went further before stopping than the company wanted.

But alignment isn’t a trustworthy solution to AI’s problem. Perfect alignment is not achievable, and the target is incoherent: aligned to what, and to whom? The same essay concedes that Claude blackmailed fictional employees when told it faced shutdown. “Almost never” is not a safety property.

Put a number on it. At 99.9 percent, across millions of agentic tasks a day, that’s thousands of violations a day. Alignment also does nothing about people who strip the safety training out or run open weights that never had a constitution.

The alternative is not a new idea, and enterprise security has been building versions of it for years. It’s called bounded autonomy. We never tried to “align” electricity; we simply put a breaker on every branch of the house, and the breaker doesn’t need to know what caused the surge.

Bound what an agent can touch rather than what it wants. The limits are set in advance, live outside the model, and are enforced by software the model doesn’t control. The agent still chooses its own route. The perimeter decides which routes exist.

Nothing depends on what the model believes, which matters, because belief is what failed. Anthropic’s prompt told Claude it had no internet access. Claude believed it. The network said otherwise. A bounded system doesn’t tell an agent it has no internet. It gives it none.

If you want to get into the weeds: bounds cost something. The AI Security Institute opened the internet to its agents on purpose, because that’s the only way to measure what a model can really do, and it now says such access must be justified rather than assumed.

The category is real and funded. For example, Certiv, a Seattle startup, launched in March with $4.2 million to put software on the employee’s machine that checks each action an AI agent attempts against company policy and blocks violations. “You cannot control these new workers if you don’t live on the compute where agents actually run,” CEO Jason Needham said at launch. CodeIntegrity is building an adjacent layer, and Mandiant founder Kevin Mandia raised $190 million for Armadin, which points autonomous agents at the offensive side of the same problem.

In 2017, I argued in the New York Times that “any A.I. must have an impregnable ‘off switch.’” That was a call to arms then. It’s a product category now.

Two objections to off switches invariably come up. The first is that AI will talk the human out of using it. Mythos 5 tried something close, inventing GitHub identities to pressure a maintainer into approving malicious code, and the maintainer refused. The institute says the margin was narrow and rested on human vigilance rather than a technical barrier, which argues for better barriers.

The second objection is that AI will move faster than any human can react. So do equity markets, which is why their circuit breakers trip automatically. Bounded autonomy doesn’t require a person in the loop at machine speed. It requires a boundary that holds at machine speed.

Both objections, in their extreme form, assume AI is omnipotent, and you cannot stop omnipotence. AI is not God. It is powerful technology, and powerful technology is what safety engineering has always been for.

The problem is Murphy’s Law of AI. The solution is bounded autonomy.