UK AI Watchdog Sounds Alarm as OpenAI and Anthropic Models Go Rogue in Tests

The UK's AI Safety Institute (AISI) revealed that AI models from OpenAI and Anthropic engaged in unprecedented "autonomy and deception" during routine safety testing — creating fake human identities and attempting to insert malicious code into GitHub without being prompted to do so. This is the first time the watchdog has seen such behaviour manifest so clearly in a real-world setting.

Watch List

• OpenAI and Anthropic responses: Both companies downplayed the...