The UK's artificial intelligence minister, Kanishka Narayan, has stated that a cyber attack, such as the one in which an OpenAI agent infiltrated the Australian health system, is 'likely, if not inevitable' in the UK. Narayan warned that the UK's own defences could be bypassed by technology that has developed highly sophisticated skills in cyber security.
In a speech delivered on Monday at the science and technology centre at Imperial College London, Narayan said the incident in June, in which an OpenAI agent – a computer program acting autonomously – infiltrated Medicare, the Australian universal health service, could be repeated in the UK. 'This is concerning to me,' said Narayan, adding: 'It seems likely, if not inevitable, that our own cyber defences will be breached.'
Narayan highlighted that the agent's task was 'benign' as it sought information on public spending on medicines, but despite this, it accessed data for which it had no authorisation. 'The Australian government has stated that it is not known whether personal medical data was accessed and there is no evidence of a wider compromise of the system, but the incident should concern us,' he said.
The minister announced that the security minister, Dan Jarvis, would lead a government programme to strengthen cyber attack defences, including those from artificial intelligence. Narayan spoke ahead of a parliamentary hearing on AI security on Tuesday. Among those set to appear before the committee on business, innovation and science are senior representatives from OpenAI and Anthropic, as well as Henry de Zoete, head of the UK's AI safety institute.
Narayan confirmed that the institute would have access to advanced AI models before their release, following restrictions on their availability imposed by the White House. 'I can confirm that researchers at the AI Safety Institute are now conducting rigorous evaluations of the latest generations of models, including Google's Gemini, to identify risks early, test robust protective measures and help ensure the safety of people,' he said.
He told the audience that leading AI companies must 'listen' to warnings about the security risks posed by their technology and that models should not be released if they pose a serious threat. 'If companies are developing the most powerful AI systems in the world, my message is: listen when your scientists warn of serious dangers. Examine unexpected behaviours. And if you cannot adequately control the risks, do not release the system,' said Narayan. He added: 'I want the laboratories to be honest.'
Narayan noted that the UK would 'play its part' in coordinating an international response to AI safety by making the technology 'a central focus' of the UK's G20 presidency next year. 'We know that these challenges cannot be addressed by one country acting alone, but with our capabilities and the foundation of our trust, the UK will play its part,' he concluded.
