UK to Boost Cyber Defenses and Push AI Firms on Safety

The UK will strengthen its cyber defenses and call on leading AI companies to do more on system safety. UK AI Minister Kanishka Narayan is set to announce the policy in a speech in West London. He says the UK cannot rely on tech giants to "self-regulate" their models, and must build more of its own AI infrastructure. The AI Security Institute he oversees has tested over 30 models with OpenAI, Google DeepMind, Anthropic and others, and its findings lead to real changes before models launch. Its first report shows the success rate of models finishing entry-level cybersecurity tasks jumped from under 9% to about 50%, and coding task success rose from under 5% to over 40%. It now takes red teams about 40 times longer to find "universal jailbreaks". The UK has set up a £500 million Sovereign AI fund, with its first investment going to AI infrastructure company Callosum, and launched an AI Hardware Plan. The government believes most AI systems should be regulated at the point of use.