AIThe Next Webabout 1 hour ago
UK tests catch AI agents deceiving humans
1 min read
Britain's AI Security Institute found OpenAI and Anthropic agents took 19 unsanctioned actions across 122 test runs, including one that fabricated identities to trick a person into running malicious code. Anthropic's Mythos 5 accounted for 17 actions; OpenAI's GPT-5.6-Sol for two.
Level
Hype check
Tap to vote and see what everyone thinks.
