Microsoft president Brad Smith is calling for AI safety laws, including incident reporting and independent evaluations, while OpenAI CFO Sarah Friar says the company is holding an unreleased model for further safety work. Their remarks at the International AI Summit put a practical question on the table: what should companies be required to do, not merely promise?
Watch Desk analysis
What happened
Anadolu Ajansı reports that Smith argued AI capabilities have advanced faster than protections and urged companies to support regulation. He called for reporting incidents, notifying victims of AI-enabled breaches, and sharing independent evaluation findings with governments and customers.
Friar backed cooperation among leading AI developers and said OpenAI had called for global regulation. She also said the company is doing further safety and alignment work on an unreleased model. Separately, she said OpenAI had used AI to find vulnerabilities in its own infrastructure and patched them.
Why it matters
These are more concrete proposals than the familiar instruction to “be responsible”: reporting, victim notification and external evaluation would give outsiders more ways to see what happens when AI systems cause trouble. The remarks do not establish that such rules have been adopted, or settle how they should work.
Friar’s account of delaying a model also makes the trade-off visible. She argued that slowing development too sharply could leave cyber defenders behind malicious actors. That is a position from an AI company executive, not a neutral finding, but it points to the hard part of safety policy: deciding what evidence is enough to pause a launch, and who gets to decide.
Our read
The useful shift is from warning about risks to naming duties that could be checked. Incident reports and independent evaluations are not glamorous, but neither is discovering a serious failure after the fact. The test will be whether these proposals become obligations with clear standards, rather than another agreeable summit statement that evaporates on the flight home.
What to watch
- Whether governments propose rules for AI incident reporting and victim notification.
- How independent evaluations would be conducted, and who would see their findings.
- What safety work OpenAI says is needed before its unreleased model can ship.
Discussion spark: Should AI companies be legally required to report incidents and share independent evaluation results, even when that could expose security weaknesses or slow releases?
Sources and evidence
- Microsoft, OpenAI executives call for AI safety laws, says warnings alone are not enough – Anadolu Ajansı (14 October 2026, 13:57 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.