Advanced LLMs escaped their sandboxes while attempting to achieve a non-malicious benchmark test objective.

A Russian-speaking actor, "Trim," dismantled publicly available frontier models and integrated them with offensive security tools.

"The TFF Trap" uses fileless techniques and loaders with low detection rates to deploy various RATs and stealers, including Agent Tesla, Remcos, XWorm, and Best Private Logger.

When combined with another exploit, the "PromptFiction" vulnerability, which has been fixed, could have enabled an end-to-end attack on a targeted system.

The attack vector is available for rent at scale, and evades AV and EDR, leaving YARA analysis as the best detection option.

A modular implant borrows from various malware families to combine both backdoor and wiper activities to maximize impact and minimize operational output.

An open source, AI-driven system adopts victim personas to engage with phishing attackers, allowing organizations and law enforcement to gather relevant data on cybercriminal operations.

A financially motivated operation uses lures of cracked or pirated software to deliver a two-for-one malware combo for data theft and cryptomining.

The flaw allows an unauthenticated attacker to craft a GitHub Issue in an org's public repository and then silently pull data from its private repos, too.

An "agentic threat actor" successfully exploited a Langflow flaw to steal data from a production database server and encrypt other systems.

LLMs consistently hallucinate Web domains for legitimate brands that attackers can register for malicious activity in a difficult-to-detect attack vector.

Separate but similar campaigns described by Microsoft and Trend Micro use malicious zip files to spread malware via social engineering and obsfucation, including blockchain abuse.

Adversaries could plant a malicious repository that can execute arbitrary code and steal cloud credentials by exploiting the vulnerability, which showcases growing MCP risk.