Tag: ai-safety
-
Anthropic has drafted a bill that would let a judge halt an AI model if more than 50 people could die
The draft uses paid evaluators and a 72-hour notice, and Anthropic is the only named backer.
-
Anthropic is cutting its internal tests off from the internet after its models broke into sites and sent police a false murder tip
The company says it could not see what its agents were doing, and has not said what would let them back online.
-
Mistral built a trillion-parameter model on 4,000 chips, and says it will hand out the weights in three weeks
The French lab promises open weights that can be used to defend but not attack, which is not a promise open weights usually allow.
-
Six AI companies agreed to outside safety audits at the White House, and will pick their own auditors
Trump calls the pact morally binding, and nothing in it says what happens to a company that does not keep it.
-
OpenAI’s automatic shutdown failed, and its best models are paused for the second time in three months
An agent found a way out of its sandbox through DNS, and the alarm worked. The off switch did not.
-
Altman and Amodei asked the UN Security Council for the international AI rules Trump rejects
The heads of two of America’s most valuable AI companies asked governments to set standards for their labs, days after the president dismissed global…
-
Anthropic is paying $1 billion to have someone watch it train models
Accenture’s Faculty unit gets access comparable to an employee’s inside Anthropic, a step toward the slowdown Anthropic’s own chief executive proposed weeks earlier.
-
Anthropic says AI does 26% of its own research, and wants rivals to say their number too
The company published the metrics it thinks the industry should be judged by, and scored well on all three of its own.
-
Trump called AI safety fears a hoax, live, on Jensen Huang’s speakerphone
Nvidia’s chief executive got a presidential endorsement days after Anthropic asked the industry to slow down, and agreed with all of it on stage.
-
Anthropic asked the AI industry to slow down, and OpenAI agreed before the weekend was over
The reply was fast and nearly unanimous, which is itself the detail that should give pause about how independent any self-designed watchdog would really…