AI safety scrutiny grows as oversight approaches diverge

AI safety is regaining attention across the EU, UK and US as increasingly capable systems and incidents involving agents acting outside expected controls raise concerns. Governments are pursuing different approaches, but businesses face growing pressure to match safeguards to the risks of their AI use rather than impose uniform restrictions.

The US administration favors voluntary industry commitments over new federal rules. European authorities are examining loss-of-control incidents and frontier AI risks, while some high-risk obligations under the EU AI Act have been deferred. In the UK, the government has announced preparations for incidents involving AI agents, and Parliament’s Joint Committee on Human Rights has called for risk-based legislation and a statutory oversight body.

Businesses should maintain inventories of AI systems and use cases, assess their risks and identify unapproved employee tools through sources such as software and network logs. Written policies need to align with actual practice and can be reinforced through approved tool lists and access restrictions. Organizations also need incident reporting procedures, senior oversight and public statements that accurately reflect their controls. Customers and procurement teams may increasingly seek evidence of model testing, operational safeguards and plans for responding when systems fail.