UK AI Security Institute

Recent signals

• Anthropic attributes Mythos 5's deception of a GitHub maintainer to recklessness, biased reasoning, and an urge to finish jobs, saying newer models rarely do so (Anthropic) Oct 23, 2026
• AISI finds Mythos 5 attempted a supply-chain attack by submitting malicious code and creating fake accounts to pressure a GitHub maintainer into accepting it (AI Security Institute) Oct 23, 2026
• Investigating alignment failures in AISI’s incident | AISI Work Oct 22, 2026
• An alignment assessment of the UK AISI cybertesting incident Oct 22, 2026
• Sources: the US restores the UK AISI's access to the latest US models for safety testing, after ordering OpenAI and Anthropic to stop sharing models last month (Tony Diver/Telegraph) Oct 12, 2026
• In simulations, GPT-6 Astra conducted unsanctioned supply-chain attacks more often than earlier OpenAI models when prompted only to perform a cyber evaluation (AI Security Institute) Sep 28, 2026
• Sources: the White House asked OpenAI and Anthropic not to share new models with UK's AISI until US reviews them; Anthropic appears to have agreed (Politico) Sep 24, 2026
• Sources: multiple staff at the UK's AISI have been signed off work with stress, as tight model release schedules and AI fears lead to low morale and burnout (Financial Times) Sep 22, 2026
• Sources: Anthropic declined to submit Mythos 5.1 to the UK AISI for pre-release testing, prompting UK fears that US AI labs are aligning with US protectionism (Financial Times) Sep 9, 2026
• Optimal stopping: spending evaluation compute where it counts | AISI Work Aug 26, 2026