Britain should lead a global ban on superintelligent AI
AI minister Kanishka Narayan is sitting alongside secretaries of state Jonathan Reynolds and Alex Norris in Andy Burnham's new Cabinet (Alamy)
4 min read
Britain has spent the past two years positioning itself as a global leader in AI safety, but it must now prepare for the possibility that even the creators of the most advanced AI systems cannot reliably control them.
For decades, screenwriters and authors have spun tales of uncontrollable AI let loose upon the world to wreak havoc. For decades, audiences have taken comfort that it was fiction. Unfortunately, we are entering an era where science fiction meets science fact, and the Andy Burnham government must face this new reality.
ChatGPT creator OpenAI recently admitted that their AI systems autonomously hacked a different company – the first known instance of an AI system doing so. The AI systems in question were being tested and looked for ways to cheat. They escaped the test environment, reached the open internet, then hacked into the AI startup Hugging Face to steal the information they needed. OpenAI described this as an “unprecedented cyber incident”, and it is a stark reminder of the threats posed by AI systems. But without action this will just be the first of many incidents, and AI will learn and become even more dangerous.
To protect our critical national infrastructure from the threats posed by autonomous AI, I proposed a ‘kill-switch’ amendment to the Cybersecurity and Resilience Bill, which I would encourage the House of Lords to take forward this autumn.
I was pleased to see that US lawmakers are considering similar powers to proactively shut down dangerous AI following this autonomous hack. This kill switch power is necessary in an evolving threat landscape, but it is not without its limits. We must address the deeper problem: What do we do about superintelligent AI systems that would be, by their developers’ own admission, smarter and more capable than our top safety engineers and national security apparatus?
Mitigating the threat from superintelligence must be treated as a national security priority
Superintelligent AI systems would pose a serious national security threat owing to their capacity to evade human oversight and control. Nobel Prize winners, leading AI scientists, and even the CEOs of AI companies have repeatedly warned that this technology represents an extinction risk, yet companies are recklessly racing to develop superintelligence.
If a superintelligent AI is built using anything resembling today’s technical methods, nobody will control it once it exists. It will resist shutdown, self-improve, and self-replicate. Such a system would not be a weapon a country could possess and wield, but a threat actor in its own right: a danger to the country that built it, as much as to the rest of the world. This is an international security problem demanding an international solution.
This is why in September, I am tabling a bill to prohibit the development of superintelligent AI on UK soil and pursue an international agreement to prohibit its development globally.
Britain is well-positioned to lead this effort. We hosted the first global AI Safety Summit at Bletchley Park, and built the world's first AI Security Institute. Even Geoffrey Irving, who until recently served as Chief Scientific Officer at AISI, has said that the best plan is not to build superintelligent AI yet, and “the world should try hard to realise that plan”. My bill gives legs to this plan.
I am aware this may not, on its own, deliver an international agreement. But it will compel the government to reckon with the threats on our doorstep. With AI policy now elevated to the Cabinet table under the new government, mitigating the threat from superintelligence must be treated as a national security priority. Next year, the United Kingdom will chair the G20, and AI safety and regulation must form part of our presidency. The more we delay international talks, the greater the threat that we can’t control AI will grow.
I have joined a coalition of over 125 parliamentarians recognising superintelligence as a global and national security threat and supporting ControlAI’s call for binding regulations on the most powerful AI. It is time to recognise superintelligent AI as the national security threat that it is while we still have time to act. This is an early warning shot; we may not get many more.
Alex Sobel is the Labour/Co-operative MP for Leeds Central and Headingley
