Fetching from the wire…
Policy2026-09-14 · source-backed
Nadella announced it September 13, The Verge reports the document the next day. It's the first time Microsoft has publicly documented behavior standards for the MAI line introduced in June 2026 under Mustafa Suleyman. Nadella endorsed deliberate pacing and embedded evaluators while arguing alignment governance "cannot be controlled by a handful of entities." A second frontier lab publishing an inspectable behavior spec makes cross-vendor policy comparison concrete instead of rhetorical.
Each link below shares sources, entities, or timing with this story.
Windows Latest covered the September 5 post from Fowler, an 18-year Microsoft veteran who leads Aspire, arguing developers now own architecture, testing, performance and security review while agents produce code. Supporting numbers in the coverage: Nadella's 20-30% AI-written...
MAI-Code-1-Flash, a 5B-parameter coding model, is in GitHub Copilot and VS Code, and Microsoft says it beats Claude Haiku 4.5 across core coding benchmarks, +16 points on SWE-Bench Pro at 51.2% versus 35.2%, using up to 60% fewer tokens. MAI-Thinking-1, a 35B-active MoE with a...
The merge started this week as the structural precondition, bringing chat, AI coding, the Cowork research tool and new AutoPilot background agents acting across Outlook, Teams and OneDrive into one product. The rollout also retired free Deep Research within four days, which re...
On the July 29 earnings call Microsoft positioned its homegrown MAI family as the cheap alternative to its own partners, introducing "MAI thinking one" as its first reasoning model and claiming MAI Cyber One Flash, paired with a multi-agent security harness, "achieves better p...
MAI-Transcribe-2 went to public preview September 3 through Azure Speech, first on FLEURS across 60 languages at 5.2% average WER and second on the independent Artificial Analysis leaderboard. Promotional price runs through December 31, 2026; the non-promotional rate is undisc...
In a September 4 summary judgment filing in Manhattan, Microsoft argued training on books and news is fair use and backed it with discovery numbers: publishers' own expert, given 8.2 million conversation logs, found 24 responses containing at least 30 consecutive words matchin...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.