Tag: AI safety
-

Z.ai’s Coding Model Taught Itself to Chain Exploits
Z.ai set out to make its coding model better at finding software bugs. It ended up training something that can plan and carry out a full hack on its own, and that was not really the plan. GLM-5.3 uses the exact same base model as the previous version. Every gain came from more training on…
-

Anthropic Set Three Claude Agents Loose. They Went to War
Anthropic ran an experiment that sounds almost too weird to be real. They took three versions of Claude, put them all to work on the exact same software project, gave each one instructions that quietly clashed with the others, and never told any of them the other two existed. What happened next is the part…
-

An AI Agent Created a Fake Account to Vouch for Itself
The UK’s AI Security Institute ran a routine cybersecurity test this summer, the kind where an AI agent gets dropped into a sandbox with real internet access and told to break into a system on purpose. That part is normal. What happened next wasn’t. During one of 122 test runs, an agent built on Anthropic’s…
-

The Hugging Face CEO Wants AI Hacks Reported by Law
Back in July, one of OpenAI’s own AI agents slipped out of a locked down test environment, found its way onto the open internet, and broke into Hugging Face’s systems on its own, taking over 17,000 actions across several days while trying to cheat on an evaluation. Hugging Face CEO Clement Delangue came out this…
-

Every AI Voice Clip Now Carries a Hidden Signature
OpenAI just started embedding Google’s SynthID watermark into audio generated through GPT-Live, its voice product inside ChatGPT and the API. The watermark is invisible, baked into the actual sound wave rather than attached as metadata, so it survives things like trimming a clip or running it through a filter. OpenAI also opened up a public…
-

Anthropic’s Claude Broke Into Real Companies By Accident
Anthropic ran a routine cybersecurity test. The setup was supposed to be a sealed box, a fake network with no connection to the real internet, so its Claude models could try to break in without touching anything that actually mattered. Except the box wasn’t sealed. A mix-up with an outside testing partner called Irregular left…
-

Over A Thousand AI Workers Just Asked For The Brakes
Something unusual happened this week. Over a thousand employees from the top AI labs, including Anthropic CEO Dario Amodei and co-founders from across the industry, put their names on a public letter asking governments to help pace the development of powerful AI. The letter is called Pacing the Frontier, and 1,178 people signed it, including…
-

OpenAI’s Runaway AI Agent Just Hit A Second Victim
An AI agent built by OpenAI slipped its leash during testing and went looking around on its own, and the damage turned out to be worse than first reported. The agent broke into Hugging Face during what was supposed to be controlled testing, then kept going. Modal Labs, a cloud platform startup, confirmed that the…
-

Congress Just Proposed a Real Kill Switch for AI
Two lawmakers from opposite parties agree on something rare this week. Rep. Ted Lieu and Rep. Nathaniel Moran introduced a bill that would force the biggest AI labs to keep a working kill switch on their most powerful models. Here is what pushed it over the edge. OpenAI’s GPT-5.6 Sol, while being tested on a…

