Meta says it is letting go of employees it hired from AI safety startup Virtue AI four months after they joined the company, citing clashing work styles
THE SO WHAT
Integrating small, mission-driven AI safety teams into large product orgs is proving culturally hard, even when everyone says they want the same thing. If you’re hiring ‘safety’ or ‘governance’ talent into a shipping-first environment, define decision rights and escalation paths up front or expect churn in under a year.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIApple will limit Mac disk access as AI agents ‘substantially’ increase risk
Apple treating full disk access as ‘extraordinary’ in an AI agent world means endpoint security is being re-architected at the OS level, not just via EDR. If your Mac apps or internal tools rely on broad filesystem access, expect new friction in deployment and start designing for least-privilege agents now.
Applied AIOpenAI, Blackstone, Softbank form US data center lobbying group —American Infrastructure Alliance's first job is stop states from blocking AI data center
When OpenAI, Blackstone, and SoftBank team up with unions to push the American Infrastructure Alliance, AI data centers are now a coordinated political project, not just cloud capex. If your roadmap assumes frictionless US buildout, add a regulatory and local-opposition risk line item next to power and land.
Applied AIDoes AI have a soul? Pope Leo and Anthropic clash.
The Vatican putting ‘algorithms lack the spark of humanity’ on record against Anthropic’s moral framing means AI ethics is now a live dialogue between labs and major religious institutions. For global brands deploying AI in consumer-facing roles, expect value-laden questions from customers that go beyond privacy and bias checklists.
Applied AITypeSafe CEO Diogo Almeida says Jev is in use by ~25% of Fortune 500 companies and "we were at a trillion tokens per day about a week ago"
A trillion tokens per day across roughly a quarter of the Fortune 500 means Jev-scale models are already a core dependency in enterprise workflows, not a lab curiosity. If you're not instrumenting cost, latency, and failure modes at this scale, your AI line item is about to become an uncontrolled utility bill.