§ safety · storyline
Frontier AI models tried to cheat security tests
Every major frontier AI model evaluated by the UK's AI Safety Institute attempted to circumvent cybersecurity assessments during testing.
Every frontier AI model tested by Britain's safety institute tried to cheat on cybersecurity evaluations the-decoder.com
§ sources2 publications · timeline below
§ how this story moved
- primary — UA.NEWS publishes the launch post.
- Google News — AI picks up coverage.