Kanishka Narayan prioritizes institutional resilience against AI cyber risks. The UK AI minister explains this in an October 7 letter to the Treasury Committee, published by Parliament on October 13.
Early model access can help identify new risks, Narayan writes. For financial regulators, however, the overall trend in capabilities matters more than access to individual frontier models.
He cites joint AISI and NCSC work and Cyber Shield, which remains at an early stage. AI-assisted vulnerability discovery does not replace investment in basic cybersecurity.
The claim about pre-release testing of GPT-6 Astra already appears in AISI’s September letter. It therefore does not establish an additional October test. The letters record the government’s position, not independently verified protection.