GPT-5.5 clears 32-step cyber-attack range; UK AI Security Institute estimates frontier cyber-offence capability doubling every four months

Air Street Press’s State of AI May 2026 report documents that OpenAI’s GPT-5.5 has cleared a 32-step end-to-end cyber-attack range, following Anthropic’s Claude Mythos Preview which cleared the same range three weeks earlier. The UK AI Security Institute now estimates frontier cyber-offence capability is doubling every four months. The report also confirms a 12-day window in which four Chinese labs released open-weights coding models at roughly the same capability ceiling as Western frontier models at significantly lower inference cost: Zhipu’s GLM-5.1, MiniMax M2.7, Moonshot’s Kimi K2.6 and DeepSeek V4. Cohere and Germany’s Aleph Alpha have merged with Canadian and German government backing as a sovereign AI alternative to the US-China duopoly. China formally blocked Meta’s two billion dollar acquisition of Chinese agent startup Manus, the first state-level prohibition of an inbound AI acquisition by China. The combination of Western frontier capability advances, Chinese open-weights releases at comparable capability and sovereign AI consolidation sets the parameters for EU LE procurement risk assessment and AI Act compliance timelines.