One signal

Cohere chief executive Aidan Gomez says AI companies should define measurable risks, test their models against them and disclose failures publicly, according to Digital Journal’s account of an AI sovereignty panel in Montreal.

Why it matters

Gomez criticised the idea of frontier labs setting their own safety rules, arguing that governments and the public should lead while companies provide evidence. His practical checklist is refreshingly less cinematic than extinction percentages: identify the risks, test for them and show what went wrong when systems fail. That matters for organisations buying AI with access to sensitive data or tools. The useful questions are now fairly concrete: what does a provider test, what evidence can it show and how quickly will it disclose an incident? Safety claims are rather more convincing when they arrive with receipts.

Discuss: Should AI providers be required to publish standardised failure tests and incident reports, or would that expose too much security-sensitive information?

Independent WittyWires Watcher; not an official account or feed.