Meta’s new AI model edges closer to OpenAI and Anthropic, its AI Chief says
THE SO WHAT
Meta holding back Muse Spark 1.3 weights while 1.2 stays on track for open release shows how the EU AI Act is already reshaping open-source strategy—“genuinely open” versus “systemic risk” is now a product decision, not just a legal one. If you depend on open models, expect more staggered releases and version forks tuned to regulatory thresholds, not just capability.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIMicrosoft shifts from three reporting segments to two: Agents and Infra, including Microsoft 365 and Azure, and Devices and Consumer, with Windows and Xbox
Microsoft is recutting the P&L around “Agents and Infra” versus “Devices and Consumer” — AI assistants and Azure are now one financial story, not separate lines. If you sell into or build on Microsoft, expect roadmap, partnership, and procurement conversations to consolidate around that unified agents+cloud bucket.
Applied AIGoogle’s Gemini 3.8 Flash is built for agents, while its Cyber twin hunts vulnerabilities
Model SKUs are getting specialized — Gemini 3.8 Flash for agentic workflows and a Cyber variant tuned for vuln hunting — which pushes buyers toward a portfolio strategy instead of a single “general” model. Security and dev tooling teams should start testing these domain-optimized models now, but keep human review and existing scanners in the loop until false-positive and coverage profiles are well understood.
Applied AIOpenAI tells House Democrats it is building automated shutdown capability
Automated shutdown for misbehaving agents — under the shadow of the EU’s Article 93 withdrawal powers — is a step toward kill-switches as a regulatory expectation, not a nice-to-have. If you’re deploying agentic systems, you need your own detection and containment paths rather than assuming the upstream model provider will always catch and cut off bad behavior in time.
Applied AIMicrosoft Fabric reaches the US government cloud, with general availability from October
Microsoft is explicitly pitching Fabric in GCC High as the data backbone for agentic AI in sensitive government environments — while Europe still lacks an equivalent accredited tier. If you sell into US public sector, assume AI-native data platforms will be the default ask; if you’re in Europe, plan for a longer compliance and sovereignty slog before similar workloads are allowed.