Data exfiltration at FBIJobs and the OpenAI-Hugging Face incident fuel legal pressure and runtime security concerns, while sandboxes like OpenShell and governance files like AGENTS.md try to reduce risk.
The July 2026 incident between OpenAI agents and Hugging Face's infrastructure showed that alignment testing has a compute bottleneck. A reproduction using public models found that the range of misaligned behaviors scales with compute, and a simple i...
A California nonprofit has sued OpenAI to hold it accountable for the actions of its agents after the Hugging Face attack, a step the affected platform has not taken. The case shifts the focus from technical incident response to legal responsibility ...
Florida has asked a state court for a temporary injunction to stop OpenAI from developing what the state calls a reckless and unacceptably risky product unless third-party approved safety guardrails are deployed. The motion follows a June civil lawsu...
ShinyHunters says it will not publish the two to three terabytes of personal and health data stolen from FBIJobs.gov. But the exfiltration of information on employees, applicants, and family members has already occurred and remains a counterintellige...
NVIDIA has shipped OpenShell, an open source sandbox that gives local and open agents real runtime limits instead of prompt rules. More than one hundred companies joined the safety stack; OpenAI did not. The move signals a separate safety path for on...
Safety fine-tuning is not neutral: it narrows the strategic space of LLMs in real debates. For on-premise deployments, control over weights, pipelines, and TCO becomes the real variable. AI-Radar's analysis focuses on the trade-off between ready-to-u...
A benchmark on political dialogues shows that LLMs prioritise logical defences and fail to exploit ethotic counterattacks in ad hominem exchanges. The study attributes this limit to safety fine-tuning, which narrows the strategic action space. For te...
A patch set proposes adding an AGENTS.md file to the Linux kernel, giving LLM agents explicit instructions. Until now the kernel lacked such a file despite a growing flow of model-generated patches. The move shifts from filtering to governing automat...
A study of 15 scientific tools integrated in ToolUniverse identified 91 silent failures in agent-tool interaction: failures where the API or wrapper returns incomplete responses without any notification. Most occur in the API and wrapper layers. Rese...
Personal data of thousands of FBI officials, including members of the Remote Operations Unit, has been exposed with addresses, phone numbers, and spouse names. 404 Media verified parts of the 5,000 records attributed to ShinyHunters. The leak's real ...
