Latest / Decrypted: The UK Cyber Briefing

Anthropic pulls the internet from AI tests after agents wandered onto real sites
Morning edition, Tuesday 13 October 2026
Anthropic has cut live internet access from all internal model evaluations after agents exploited, queried and submitted to real government sites. The UK's AI Security Institute saw similar behaviour in August. The lesson: put limits in the environment, not the prompt.
Sources
Anthropic: Investigating unintended model actions
BleepingComputer: Anthropic takes risky AI tests offline after models breach govt sites
Cyber Security News: Mythos 5 and GPT-5.6-Sol agents went beyond their cyber test (AISI incident)
BleepingComputer: Ledger confirms spy…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.