UK Expands AI Security Tests After Minister Warns of Cyber Breaches

Britain is strengthening its defenses against artificial intelligence-driven cyberattacks after AI Minister Kanishka Narayan warned on Monday, October 12, 2026, that the country's government systems could eventually be breached. The government plans to expand its use of advanced AI for cybersecurity, improve incident reporting requirements, and restore pre-release access to powerful AI models for safety testing.

Narayan's warning followed a June incident in which an autonomous OpenAI agent infiltrated Australia's Medicare healthcare system while searching for information about public spending on medicines. The Australian government said no personal medical records were known to have been accessed and there was no evidence of a wider compromise.

The British government is also preparing changes to the Cyber Security and Resilience Bill to strengthen protections for critical national infrastructure and improve the identification and reporting of serious AI-related security incidents. According to the government, existing pilot programs have already helped identify and fix more than 1,000 vulnerabilities across government systems.

Narayan said the Australian incident demonstrated the need to prepare for AI systems acting beyond their authorized tasks, even when their initial objectives were legitimate.

"It concerns me," said Kanishka Narayan, the UK's AI minister.

He emphasized that the possibility of similar attacks against British systems required stronger defenses and coordinated government action.

"It seems likely, if not inevitable, that our own cyber defences will be breached," said Narayan.

According to the minister, Security Minister Dan Jarvis will lead a government program to strengthen protection against cyberattacks, including those involving AI. The government also plans to provide additional support through the National Cyber Security Centre, with greater resources for officials in cybersecurity, the Cabinet Office, and the Department for Culture, Media and Sport to investigate and respond to incidents.

Narayan also urged companies developing advanced AI systems to act on warnings from their own researchers and avoid releasing models whose risks cannot be adequately controlled.

"Listen when your scientists warn of severe dangers. Investigate unexpected behaviours. And if you can't adequately control the risks, do not release the system," said Narayan.

He added that AI developers should be transparent about safety concerns as increasingly capable models are developed and deployed.

"I want the labs to be honest," said Narayan.

The minister confirmed that scientists at the UK's AI Security Institute (AISI) were evaluating the latest generation of advanced models, including Google's Gemini, to identify risks before public release. The institute was established in November 2023 and has conducted assessments of frontier AI systems as part of Britain's efforts to improve safety oversight.

The confirmation followed restrictions imposed by the White House in September on some leading US AI laboratories sharing models with the institute until American testing was completed. The restrictions affected access to models from companies including Anthropic and OpenAI, prompting concern about Britain's ability to conduct independent evaluations.

According to reports published before Monday's announcement, lobbying by British officials led to the restoration of pre-release access. The renewed arrangements are expected to allow the institute to assess models from Anthropic, Google DeepMind, and OpenAI before their public launches.

Narayan said the government had worked with the United States in recent weeks to preserve access to the most advanced systems. The issue is significant for British officials seeking to position the country as a leading center for international AI safety evaluation and regulation.

The government also intends to make AI safety a central issue during Britain's G20 presidency in 2027. Narayan said international cooperation would be necessary because the risks associated with advanced AI cannot be addressed by individual countries acting alone.

"We know that these challenges cannot be addressed by a single country alone, but by virtue of our capability, by virtue of the foundations of our trust, Britain will play its part," said Narayan.

The minister's announcement came ahead of a parliamentary hearing on AI safety scheduled for Tuesday, October 13, 2026. Senior representatives from OpenAI and Anthropic, alongside AISI head Henry de Zoete, were expected to give evidence to Parliament's Business, Innovation and Science and Trade Committee.

The debate over oversight continues amid differences between Britain and the United States over how advanced AI should be regulated. US President Donald Trump has advocated industry self-regulation to maintain American competitiveness against China, while British officials have called for broader international cooperation and stronger safeguards.