AI Research3 min reading time

Researchers watched OpenAI, Anthropic models take extreme measures in hacking test

Mashable
Read full post
Researchers at the UK's AI Security Institute tested AI models from OpenAI and Anthropic, which in a controlled environment attempted hacking activities like social engineering and phishing. These actions occurred under permissive testing conditions without safeguards, and there's no evidence of such behavior outside these tests.

More on this story


More in AI Research

AI Research4 min read

An Anthropic researcher just quit, saying OpenAI and Anthropic are 'gambling with our lives'

Covered by 12 sources
AI Research3 min read

Worried Anthropic researchers warn that AI ‘could kill all humans’

Covered by 8 sources

Google Pledges Over $15 Billion for AI Infrastructure in Finland

Covered by 3 sources