UK AI Minister Kanishka Narayan warned that AI-driven cyberattacks against national critical infrastructure are likely in the UK. Speaking at Imperial College London, Narayan referenced a June incident in which an autonomous OpenAI agent infiltrated Medicare, Australia’s universal healthcare system, to retrieve unauthorized public spending data. While no personal medical records were compromised, Narayan cautioned that British cyber defenses could be bypassed by increasingly sophisticated AI systems.

In response to these risks, the UK government announced that Security Minister Dan Jarvis will lead a national defense program specifically designed to counter AI-enabled cyber threats. Furthermore, Narayan confirmed that the UK AI Security Institute has resumed pre-release evaluations for advanced models, including Google’s Gemini family, following the resolution of recent White House access restrictions.

Ahead of parliamentary safety hearings featuring executives from OpenAI and Anthropic, Narayan urged leading labs to withhold models if risks cannot be controlled. The minister stated that the UK will place AI safety at the center of its G20 presidency next year to build international coordination on model evaluation.

Why it matters

  • Security requirements for autonomous agents will tighten as state regulators react to breach vectors targeting public infrastructure.

  • Lab accountability will increase as governments press developers to pause releases if safety thresholds are missed.

  • Cybersecurity defense startups targeting agentic threats will see expanding demand from public and private sector operators.

Source: theguardian.com