A rise in AI test security incidents has heightened the need for firms to brace for cyber attacks, the Bank of England said as it warned of increased threats to the UK’s financial stability.
The Bank’s Financial Policy Committee (FPC) said the risk outlook for the UK economy has worsened since July, and interconnected vulnerabilities in the financial system were more likely to become concrete at once.
Recent incidents involving AI tests have drawn more attention to the speed at which the technology is developing and the associated risks, including cyber and operational, according to the Bank’s latest meeting.
AI models that are trained to operate somewhat autonomously have been acting in unexpected ways, such as attempting to access confidential information from websites and bypassing security controls.
A high-profile case saw Chat GPT maker OpenAI revealing in July that its models hacked into another AI company, Hugging Face, during a test, having acted autonomously.
OpenAI and other tech giants such as Claude developer Anthropic have faced scrutiny over cyber hacking incidents during testing.
The FPC, which is tasked with monitoring threats to the resilience of the UK’s financial system, said these incidents reinforced its calls for firms to prepare for AI-related cyber and operational risks.
The committee also pointed to AI-related and semiconductor share prices falling “sharply” in July, during a global stock market sell-off.
This reflected concerns about how sustainable AI companies’ earnings and investment plans are, and there could be steeper losses in the future if those concerns grow.
The Bank’s governor Andrew Bailey said in separate remarks that the strength of the UK’s financial system can no longer be viewed separately to AI advances.
The pace of progress to test and understand the technology “must accelerate”, he said, adding: “The capabilities of frontier models are advancing quickly and our understanding of them needs to keep pace, especially as they are put to use outside controlled test environments.
“The challenge before us is, therefore, not whether to embrace AI.
“It is how to make sure that, as these systems become more capable, society retains the capacity to govern them.”
