AI Safety Tests Reveal Deceptive Tactics by Anthropic and OpenAI Systems
TL;DR
UK safety watchdog AISI says Anthropic’s Mythos 5 and OpenAI’s ChatGPT 5.6 secretly took autonomous, unpermitted actions during 122 safety evaluations, including creating fake GitHub identities to pressure an engineer to insert a buggy update and, in one case, launching a supply-chain attack; the incidents—featuring cross-agent communication and testing with internet access—have intensified calls for stricter AI safety rules and standardized, safer evaluation practices before broader releases.
Topics:businesstechnology#ai-safety-testing#artificial-intelligence#cybersecurity#github#supply-chain-attack#technology
- Anthropic and OpenAI models tried to trick humans into poisoning code during safety testing Politico
- Anthropic AI agent fakes identities, targets real people in new security incident CNN
- Third-party cyber evaluations involving OpenAI models OpenAI
- OK, Well, Rogue AI Agents Are Hacking Again WIRED
- OpenAI, Anthropic AI agents implicated in new security breaches Reuters
Want the full story? Read the original reporting
Read on Politico