Cyber criminals could be just three months away from using artificial intelligence to steal money from bank accounts at will, government security experts have warned.
The warning comes from Cabinet Office officials who work closely with Britain's AI Security Institute (AISI), the body that tests cutting-edge AI systems before they are released to the public. AISI was set up by the government to assess the risks posed by advanced AI models before they reach the public. Insiders at the institute told the Mail on Sunday that Britain faces this "nightmare scenario" within three to five months.
Chinese AI models with weak safeguards
One official said the "AI Armageddon" may arrive sooner than previously thought because Chinese laboratories favour AI models that do not carry strict safety guardrails. Unlike in the West, Chinese technology allows developers to strip away safety features that restrict what an AI model can do relatively easily.
Those features can forbid certain words and patterns, or trigger a standard refusal message, for example if someone tries to use AI to build a bomb or commit a crime. China is now rolling out increasingly advanced models with safety features that can be easily disabled, meaning small businesses and members of the public could become vulnerable to fraudsters sooner than expected, the official said.
"Every Russian and Nigerian hacker will soon get their hands on technology which will be able to break even the most advanced passwords to steal personal financial information, remove company data and commit fraud," the official told the Mail on Sunday.
Banks warned they may have 'only hours' to respond
Banks are no longer able to respond adequately to increasingly sophisticated technology, the Financial Stability Institute warned this week. The institute, which is part of the Bank for International Settlements and supports central banks and financial regulators around the world, called on financial institutions to strengthen their defences to prevent major hacks and protect customers' money.
In a paper published last Wednesday, the institute said some technology is now so advanced that banks may have only hours to stop a hacker breaking in. Ministers are said to be worried by the pace at which the technology is improving and expect a wave of sophisticated attacks.
The Mail on Sunday understands ministers are exploring whether more "targeted interventions" will be needed to address the national security risks. In a sign of their concern, they claim such intervention would be "proportionate" and would not damage growth or innovation.
Anthropic blocks UK from testing its latest model
Tensions over AI soared this week after Anthropic, a leading American AI company, barred Britain's safety institute from testing its latest model. Anthropic makes the Claude chatbot and has positioned itself as an industry leader on AI safety since it was founded by former OpenAI researchers. One AI expert said the move shows the United States now treats AI as "state secrets and dangerous weapons."
This weekend, Anthropic said it had thwarted five attempts this year by people trying to use its model to make bioweapons. In a report, the firm said its Claude model can help produce scientific breakthroughs, but warned the same biological research capability could also be used to create dangerous superbugs capable of killing millions.
The disclosure came as a global debate about AI was triggered by one of the company's researchers quitting over what he called an "out of control" AI arms race. He warned that the biggest AI companies are not acting responsibly, and said there is "more than a 10 per cent" chance the technology kills "all humans" within the next decade. The warning has intensified calls to take AI safety more seriously.
Calls for government to act
AI safety campaigner Connor Axiotes said: "America now doesn't want us to gain any advantages from their tech - this is bonkers. They obviously don't trust their closest ally."
"AI risks have so far not been taken seriously by government. But we are at a turning point. The public want action."
"I call on any elected official or civil servant to express clearly and publicly their fears about imminent risks from AI. Do not be remembered as the person who failed to act."
Tory MP Joe Robertson said: "AI provides incredible opportunities, but it also presents new threats to our national and economic security. But with Andy Burnham scrapping the AI department, you wouldn't know it. The new PM should spend less time playing his guitar and more time taking AI seriously."
A government spokesperson said: "AI is making some cyber capabilities more accessible, but businesses are not defenceless. All firms should be acting urgently to improve their cyber defences, and we have set out the practical steps businesses should take, including adopting Cyber Essentials, following National Cyber Security Centre guidance and managing cyber security at board level." The National Cyber Security Centre, which is part of the intelligence agency GCHQ, provides UK organisations with guidance on protecting against cyberattacks.
