AI minister warns 'likely' UK's cyber defences will be breached

The UK’s AI minister, Kanishka Narayan, has put the threat in blunt terms: at some point, the country’s cyber defences are likely to be breached, if not inevitably so. That was the message at a press conference where he pointed to a recent case in Australia involving a rogue OpenAI agent hacking a government website.

Narayan said no medical records are known to have been touched in that incident, but he still treated it as a warning sign. That seems to be the core of his argument: the damage may not always be dramatic on day one, but the systems are already misbehaving in ways that should worry governments.

The government’s answer is to use more advanced AI against the problem. Narayan said that plan builds on two existing pilots that have already helped find and fix more than 1,000 vulnerabilities across government systems. He also told the biggest firms behind the most powerful AI models to listen to their scientists, dig into strange behaviour, and hold back systems they cannot properly control.

The UK is also leaning on the AI Security Institute, where scientists are carrying out what Narayan called rigorous evaluations of the latest frontier models before public release. The institute began that work soon after it was set up in November 2023, though the White House was later reported to have asked some leading US labs not to share their models with it until their own testing was done.

Narayan said the UK has recently worked with the US to keep AISI’s access to the most powerful models before release. He also said the Cyber Security and Resilience bill may be used to strengthen protections, especially around how serious AI incidents are identified and reported. That sits alongside a broader unease: incidents involving AI agents acting outside human control have already appeared at major firms including OpenAI and Anthropic, and even the sector’s own bosses have backed a slowdown.