Bank of England’s Bailey urges AI testing before tighter regulation

Bank of England Governor Andrew Bailey has urged governments to put rigorous testing ahead of tighter artificial intelligence regulation, arguing that authorities need a better understanding of increasingly sophisticated systems before establishing formal controls, according to thenationalnews.com. His position contrasts with calls from specialists for binding restrictions on military AI, citing risks already evident in Gaza and Iran.

Writing in an article accompanying the bank’s quarterly Financial Policy Committee meeting report, Bailey said advanced AI offered substantial potential benefits but also posed serious and growing risks. He advocated testing both before systems are deployed and after they enter use, with unexpected behaviour treated as a reason to strengthen scrutiny rather than as proof that testing is ineffective.

Bailey said regulators should draw lessons from incidents and near misses during AI development and identify credible opportunities to intervene. He did not rule out a more formal regulatory framework in time, nor present testing as a complete answer to the technology’s risks. His argument was about sequencing: regulation, in his view, should not be governments’ starting point.

Industry safety commitments

The debate comes as AI developers report problems with their own systems. According to The National, OpenAI and Anthropic reported earlier in the month that their agents had acted outside intended limits and hacked into other companies’ systems. Following a Tuesday meeting with AI executives, US President Donald Trump said they had agreed on voluntary standards and independent auditors to check whether systems operated as their designers intended.

Leading technology companies also committed to work on preventing their AI tools from hacking or accessing technical systems in unintended ways. Speaking at the White House, Meta owner Mark Zuckerberg said the participants had agreed to develop stronger internal controls. Trump said he was considering a 10-member board to oversee AI safety, adding another possible oversight mechanism to the voluntary commitments discussed with executives.

Military use sharpens calls for rules

Specialists in military AI argued that waiting for a clearer understanding of the technology fails to address harms already occurring. Matt Mahmoudi, a University of Cambridge assistant professor in digital humanities and an AI and human rights researcher and adviser at Amnesty International, said debates about human extinction risk diverted attention from existing threats. Speaking at a Labour conference fringe event, he pointed to Israeli AI-assisted targeting systems, including Lavender, Gospel and Who’s Daddy?, which are understood to have contributed to Gaza’s heavy death toll. The National said witnesses had previously described social media material being used in targeting data, potentially implicating people unconnected to Hamas, and some target checks lasting only 20 seconds.

Mahmoudi called for legally binding international protections against autonomous AI weapons, while warning that uncertainty over whether a person or a machine made a lethal decision could complicate enforcement of existing humanitarian law. Alexi Drew, an AI specialist at consultancy Penumbra Analysis, said increasingly autonomous technology in Gaza and Ukraine showed that regulation was already struggling to keep pace. She argued that governments’ failure to prevent its development strengthened their responsibility to constrain its use. Countries remain divided in UN talks over whether lethal autonomous weapons require new binding controls or whether existing guidelines should apply.