Microsoft says its AI models will never resist being shut down
THE SO WHAT
Committing that in-house models will “never resist human interruption” moves shutdown behavior from an implementation detail to an explicit governance promise. When you evaluate AI vendors, start asking for similarly concrete behavioral commitments — and how they’re technically enforced and audited.
READ THE SOURCE
MORE FROM THE WIRE
Applied AITom Duff Gordon, OpenAI's EMEA policy head, calls on the UK to introduce binding AI rules to "establish durable, mandatory, capability-based requirements"
A major lab explicitly asking the UK for binding, capability-based rules is a push toward regulation keyed to model power, not use case. If you sell into regulated sectors in the UK or EU, assume future compliance will hinge on the capabilities of the models you use and log those dependencies now.
Applied AIMicrosoft’s new AI rulebook rejects the research its $5bn partner is doing
Microsoft’s AI code of conduct drawing a hard line against model welfare or rights while a $5B partner explores that question publicly shows governance is fragmenting even inside tight alliances. For enterprises, this is a reminder to pin your own red lines in writing rather than inheriting a vendor’s philosophy by default.
Applied AI‘It 100% sounded just like her’: AI voice scams are getting frighteningly convincing — here’s how to protect your family
Hyper-realistic voice cloning means “I recognize the voice” is no longer an authentication factor. Any organization with phone-based workflows — from customer support to trading — should add out-of-band verification and train staff to treat voice as untrusted input.
Applied AIBernie Sanders to Join Forces With Steve Bannon and Anthropic Researcher at ‘Pro-Human’ Conference
When a US senator, a populist strategist, and a frontier lab researcher share a stage on AI, the issue has crossed partisan and ideological boundaries. Expect more mixed coalitions around “pro-human” framing — and plan for AI policy risk that doesn’t map cleanly to left/right assumptions.