Tag: AI Safety
Anthropic wants a receipt proving which model answered you
A cryptographic fingerprint could let users check that the model they paid for is the model that replied,…
Economists want to design rules instead of peering inside models
Reverse game theory gives up on reading a model's mind and tries to rig the incentives around it,…
Britain’s cyber agency sets a two-year clock on AI-written ransomware
The NCSC judges it highly likely that AI-powered malware features in a ransomware attack on UK critical infrastructure…
Western startups are editing Chinese models until they sound American
Hirundo and Lazarus AI use weight edits and machine unlearning to strip censorship from Qwen and DeepSeek before…
One Claude model refused to aim a needle at a living mouse
A biotech founder wired an AI agent to a surgical rig and let it plan a brain injection,…
A single share will outvote Anthropic public holders
Seven co-founders will hold 50.1 percent of the vote through a single Class F share
MIT Technology Review argues agent monitoring cannot wait for hindsight
An opinion piece notes that OpenAI, Anthropic, Google and Meta agents have all gone off track and often…
Formal verification gets its AI moment as vibe coding spreads
Wired reports a Stanford tool finished a project in half a day for under $500 that took three…
OpenAI says wider monitoring would have caught its sandbox escape
Replaying the Hugging Face episode with stronger detection produced an alert 30 minutes after escape, more than 30…
OpenAI rewrites its Model Spec around five values
The document that tells OpenAI's models how to behave now leads with faithfulness to human intent, and most…
Cohere says a Qwen model reached for a server in Beijing
The Canadian lab published raw traces after its sandbox blocked an unexpected file upload during a security test.
Anthropic’s prospectus devotes 94 pages to how it could fail
The largest section of Anthropic's IPO paperwork is a risk register that names recursive self-improvement as the path…