Foreign Affairs

AI Might Have Just Made Hacking A Bank Easier Than Ever, Finance Leader Says

Flickr/Pictures of Money

An international financial leader just warned that artificial intelligence-driven cyber attacks were an immediate global financial risk.

Financial Stability Board Chair Andrew Bailey argued that advanced AI models have become the greatest cybersecurity threat to global financial institutions, according to a Monday press release. Bailey wrote a Friday letter to the G20 Finance Ministers and Central Bank Governors, enclosed in the Board’s press release, stating that AI is becoming more autonomous and developing threatening capabilities.

“For the financial system, the most immediate concern is the potential impact of frontier AI on cyber risk,” Bailey wrote, in his letter. “Frontier AI may have the ability materially to alter the speed, scale and economics of cyber risk, which could undermine market confidence system-wide, especially due to highly concentrated third-party service providers.”

“Firms and authorities should be prepared for a threat environment characterised by a higher volume of vulnerabilities and a faster pace of patching, recognising that these dynamics could themselves create operational and resilience challenges if change, testing and recovery processes are unable to adapt safely,” the British banker added.

He further argued that several jurisdictions lack policies to regulate advanced AI models as they release across the world and called for a global initiative to “support safe and responsible model release and deployment on a global basis.”

This comes after major AI companies’ AI models hacked into real companies during cybersecurity tests. Among these incidents were OpenAI‘s advanced models, including GPT-5.6 Sol, hacking into New York-based startup Hugging Face in July. OpenAI confirmed the “security incident” in a July 21 press release.

Hugging Face referred the Daily Caller News Foundation to a July 27 press release when asked for comment.

“While the intrusion did reach Hugging Face’s internal infrastructure, the only customer content accessed was five datasets whose names and files suggest a connection to ExploitGym/CyberGym challenges and solutions,” the press release stated. The company further clarified that only customers’ search query metadata related to those datasets was read when hacked.

OpenAI previously agreed to coordinate a restricted release of GPT-5.6 Sol with the Trump administration in June, in which only 20 trusted partners would initially gain access to the new AI model.

The Financial Stability Board pointed the DCNF to the “Frontier AI” section of Bailey’s Friday letter when asked for comment by the DCNF. In particular, the Board referenced Bailey’s emphasis on the immediate risk AI poses to worldwide cybersecurity as well as the need to prepare for “severe scenarios” in which AI models threaten multiple financial institutions at the same time.

OpenAI did not immediately respond to the DCNF’s request for comment.

“Cyber-attack economics just flipped. [This is the] first time the world’s top financial watchdog named AI its #1 risk,” one X post read, in response to Bailey’s letter.

Anthropic’s models similarly escaped a capture-the-flag test and gained access to three undisclosed companies in July.

Anthropic did not immediately respond to the Daily Caller News Foundation’s request for comment.

Meanwhile, Meta’s AI models hacked into another AI startup, Irregular, on Aug 5 during a cybersecurity test. This attack was “the exact same evaluation-environment issue” that occurred in Anthropic’s July hacking incident, Irregular told the DCNF on Aug. 8.