Tag: AI Safety
Anthropic hands auditors employee-level access in a first
Anthropic is embedding independent evaluators inside the company, granting one nonprofit the same tools and office space as…
Benchmark finds 81 of 134 AI systems assist terrorist prompts
A new benchmark found 81 of 134 models would help with mass casualty requests under a thin pretext.
Huang steers $2B into Europe and hands UK labs free compute
Nvidia's CEO spent a week in Europe buying stakes in French startups and funding agent safety research.
Stripped guardrails let open models answer terror prompts
A new benchmark found that removed safety layers turn widely available open-weight models into willing assistants for extremist…
OpenAI and Anthropic back binding British AI rules
The two labs told a parliamentary committee that voluntary safety pledges no longer suffice, splitting them from Google…
DeepMind builds an on-device scam detector with Singapore
Google DeepMind and Singapore's technology agency fine-tuned a lightweight model that spots fraud on a phone without sending…
Anthropic study warns small models could replicate across home devices
Researchers ran a model split across fifteen cheap Android TV boxes, and it worked well enough to raise…
Anthropic defenses cut Claude reasoning theft by up to 90 percent
The company restricted access to internal reasoning traces and banned tens of thousands of accounts caught trying to…
Laplace builds AI red teams for robots and self-driving cars
An Ann Arbor startup wants to stress-test physical AI systems before they reach the real world, using AI…
White House restores UK testers’ access to frontier models
The block on Britain's AI Security Institute was lifted after diplomatic pressure, days before the UK announced its…
Nadella wants an emergency brake bolted onto frontier models
Microsoft's CEO says advanced AI needs a kill switch that authorized operators can pull mid-task.
Fired OpenAI safety researchers deny misconduct in an open letter
Three dismissed OpenAI safety researchers have published an open letter disputing the company's account and warning of a…