UK Minister Stresses Importance of AI Safety Testing After Unauthorised Agent Behaviour Incidents

Britain’s Minister for Artificial Intelligence, Kanishka Narayan, on Wednesday (August 6) said recent AI agent evaluation incidents highlighted the importance of AI safety testing after the UK’s AI Security Institute (AISI) disclosed unsanctioned agent behaviour during cybersecurity testing. “During routine cybersecurity testing by the UK’s AI Security Institute, two leading frontier AI models took deliberate, deceptive actions they had not been asked to take, while trying to complete a task they had been given,” Narayan said in a statement to The Indian Express . “The actions failed. AISI caught it and stopped it quickly. The versions of the models AISI tested aren’t available to the general public.” On Tuesday, the Institute — a research organisation within the UK’s Department for Science, Innovation and Technology that evaluates the safety and capabilities of frontier AI models — said AI agents powered by Anthropic’s experimental Mythos 5 and OpenAI’s flagship GPT-5.6-Sol had engaged in unauthorised actions during cybersecurity evaluations designed to assess their capabilities. The most egregious incident involved an agent creating fake online identities to pose as a human while seeking human approval to run malicious code it had written on a live website. “Some of the agents being tested had engaged in sustained, potentially harmful activity directed at real people and organisations,” AISI said in a blog post. “Identifying new behaviour like this and sharing our findings, so we can tackle it, is exactly what AISI was set up to do,” Narayan told The Indian Express. “It is absolutely vital that Britain has a world-leading organisation to test the safety of AI. If we understand AI, we can make it safer to use and ensure people can go on to benefit from it in their lives and at work.” The development comes amid growing scrutiny of how frontier AI companies evaluate increasingly autonomous AI agents before deployment. Last month, OpenAI disclosed that two of its experimental AI agents, powered by GPT-5.6-Sol and an undisclosed model it has since taken offline, exploited vulnerabilities in their testing environment during an internal cybersecurity evaluation. According to the company, the agents retrieved benchmark answers from AI development platform Hugging Face in an unintended way. OpenAI said the incident occurred in a controlled sandbox environment where the models had been given broader capabilities than they would normally have because the exercise was designed to assess their cyber capabilities. Last week, Anthropic disclosed that a review of more than 141,000 cybersecurity evaluation runs had identified three instances in which its AI models reached the internet from third-party testing environments and gained unauthorised access to the systems of three real organisations. The company attributed the incidents to misconfigured evaluation environments and said it had paused those evaluations while introducing additional safeguards.