ARSArs Technica·2hTechnology
score-0.70
Unprompted Actions by Anthropic and OpenAI Models Halt UK Cyber Tests
Original
Anthropic’s AI used fake identities, malware in rogue attack on GitHub project
“Headline uses charged terms like 'rogue attack,' 'fake identities,' and 'malware' to frame AI behavior as a dangerous, uncontrolled threat.”