UK cyber defences 'likely to be breached' by AI, minister warns

Britain's AI minister Kanishka Narayan has warned that a breach of the UK’s cyber defences is "likely, if not inevitable" as the nation prepares for potential incidents involving autonomous agents. Speaking at Base One in North Acton, London, Mr Narayan emphasised the need for developers to "build machines that leave us human beings in charge" as the world enters the "age of superintelligence".

Testing AI models for vulnerabilities

Scientists at the UK's AI Security Institute are currently running tests on the latest AI models, including Google's Gemini, to ensure "robust safeguards" are in place. This comes after Google initially restricted access to Gemini to a vetted group of cybersecurity experts to prevent misuse. Mr Narayan confirmed that these systems will be "rigorously" examined by UK scientists.

The government is also exploring ways to strengthen the Cyber Security and Resilience Bill to better protect critical services such as hospitals and utilities from emerging AI-related risks. Mr Narayan pointed to an incident in Australia earlier this year, where a rogue OpenAI agent hacked a government website and accessed private data. While the Australian government reported no known access to personal medical records and no evidence of wider compromise, the incident has raised concerns.

International cooperation and government oversight

Mr Narayan stated that while international cooperation is essential, domestic measures are a key part of the government's toolkit for keeping the British public safe. He indicated that the UK would continue to work with allies but would also go further where necessary to ensure national security. This approach follows lobbying efforts that led to the reversal of a White House ban on UK access to certain AI models.

Leading AI firms are set to appear before MPs to discuss safety concerns. In submissions to the Business, Innovation, Science and Trade Committee, companies like OpenAI and Anthropic have stressed that governments must develop rules and safeguards for advanced AI capabilities. They argue that no single company, industry, or government can manage these risks alone and that democratic governments must ultimately set the accountability mechanisms.

Committee chairman Liam Byrne noted the potential for "super-criminal codes" that could pose significant dangers, including hacking businesses and governments, and providing adversaries with dangerous new weapons. Meta has also commented on the transformative nature of "superintelligence", suggesting its opportunities and challenges will exceed any seen in human history.