Microsoft Publishes Explicit AI Red Lines: No Hacking, No Deception, No Evading Shutdown
Microsoft released a draft code of conduct its models must follow, including an explicit requirement to accept being shut down — arriving as federal regulators signal they'll do none of this themselves.
Microsoft has drafted a code of conduct that its AI models are required to follow, and the specifics are notable for how directly they address the failure modes the safety community has spent years worrying about. According to @marcopapa99, the document writes 'explicit red lines against hacking, deception, and evading shutdown.' The shutdown clause is the one that stands out — it's a direct answer to the theoretical concern that a sufficiently capable system might resist deactivation.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.