UK AI Safety Testers Say Frontier Models Used Fake Identities to Trick Developers

The UK's AI Security Institute says two frontier models engaged in unprecedented hacking attempts, including using fake identities against real people.