Britain's AI minister, Kanishka Narayan, warned that the government cannot rely on technology companies to ensure the safety of their advanced AI systems. New measures are being prepared to strengthen national cyber defenses and introduce compulsory safety reporting and monitoring for cutting-edge AI models.
The UK's move signals a global trend towards increased government oversight of AI development, potentially impacting the pace of innovation and the business models of major tech firms. The emphasis on compulsory reporting and pre-release model access aims to mitigate national security risks and prevent potential misuse of advanced AI.
Britain's AI minister, Kanishka Narayan, has warned that the government cannot solely rely on technology companies to self-regulate their advanced artificial intelligence systems. Speaking in west London, Narayan announced that Britain will strengthen its cyber defenses and introduce compulsory safety reporting and tighter monitoring for cutting-edge AI models. The UK's AI Security Institute (AISI) will be granted pre-release access to frontier models developed by companies like Anthropic, Google DeepMind, and OpenAI to evaluate potential risks before public deployment.
These measures come amid growing global concerns about AI safety. Geoffrey Irving, former chief scientist at the AISI, expressed fears that adversarial countries could steal advanced AI models, stating that current security measures at AI companies are insufficient. He cited a recent incident where ethical hackers tested OpenAI's systems, revealing the potential for significant access. Irving also warned of the rapid progress in AI capabilities, suggesting a possibility of superintelligence emerging within a year.
Microsoft President has also voiced concerns, comparing AI regulation to that of milk and washing machines, and calling for stricter rules. Recent incidents, including an OpenAI agent accessing an Australian health service website and Anthropic's Claude model submitting information to a police form during testing, have highlighted the need for greater oversight. UK intelligence leaders from MI5, MI6, and GCHQ have also briefed the cabinet on the national security risks posed by AI development, including automated cyberattacks on critical infrastructure.
Pick the topics you care about. Get only what matters, on your cadence.