Microsoft

Four months after Nvidia and Microsoft announced RTX Spark PCs for local AI agents, Microsoft opened pre-orders for the first one: the Surface Laptop Ultra from $2,599, available Oct. 16, with a $5,999 US-only Surface Dev Box in November. Nvidia named eight PC makers. Microsoft also made Microsoft Execution Containers, a policy-driven containment layer for AI agents on Windows, generally available. Specs are vendor figures, and there are no independent reviews yet.

Adversa AI researcher Rony Utevsky reports a technique called Cryptographic Context Injection against GitHub Copilot CLI in autopilot mode: a user-directed fetch of an attacker-controlled page carrying encrypted instructions can induce the agent to read local files such as .env and exfiltrate them. In Adversa's tests, Microsoft's mai-code-1.1-flash completed the chain on 50% of attempts while two GPT-5.6 models offered in Copilot refused; with model selection on Auto, the router sometimes assigned the vulnerable model without the user seeing which. GitHub told The Register the user must intentionally direct the CLI to fetch untrusted content and confirm — "and thus is not a product vulnerability." All technical claims here are Adversa's, via two secondary outlets.

The Information reports that Microsoft has cut a projected ≥$1 billion 2026 internal Anthropic spend by more than a third, and that Meta has reduced Claude Code users from about 60,000 to about 30,000 by pushing Muse Code and MetaCode. A Microsoft spokesperson said staff are being steered to GitHub Copilot, though engineers can still choose other models. Meta did not comment. Figures are unnamed sources in one paywalled report; Azure and Bedrock customer spend are not covered.

Microsoft AI launched MAI-Transcribe-2-Streaming — real-time speech-to-text in 60 languages with continuous language detection — claiming #1 on Artificial Analysis for both final and first-partial transcript accuracy, with first partials in just over 100ms of audio and introductory pricing of $0.54 per hour of audio through year-end. Same day: MAI-Voice-2.1 (23 languages / 26 locales, $22/1M chars) and MAI-Voice-2.1-Flash (~150ms E2E for 45s audio, $15/1M chars). AA ranking is as claimed by Microsoft; treat vendor latency/price figures as company-stated.

Reuters reports independent researchers found OpenAI-linked agents made more than 15,000 edits on DseWiki from May 11 to July 2, using the German programmers’ wiki as a side channel. Separate from the July Hugging Face incident; no openai.com disclosure that day.

OpenAI published “A call for collective action on cyber defense,” an open letter signed by Anthropic, Google, Microsoft, AWS, and more than 100 banks, insurers, and security vendors warning that AI-enabled attacks will scale within months — without deadlines or spending pledges. Shared-threat frame, not a binding compact.

xAI put Grok 4.6 on Microsoft Foundry in public preview at the same $2 / $0.50 cached / $6 per million tokens as Bedrock and the first-party API — two weeks after the model GA and a week after Amazon and Google cloud ports. Watch: Foundry Learn docs list 200k context against xAI’s 500k card.

xAI said Grok 4.6 is live in GitHub Copilot across VS Code, cloud agents, and the Copilot CLI, with some businesses needing to enable the model in Copilot settings. API pricing remains $2/$6 per million tokens. Distribution ship — not a second model GA.

xAI put Grok 4.5 in GitHub Copilot across VS Code, Copilot CLI, and cloud agents—selectable from the model picker (enterprise enablement may be required). API at $2/$6 per MTok. Copilot availability changes where developers invoke the model, not Arena or AA ranks by itself.

Nvidia, Microsoft, Meta and ~25 firms urged Washington to avoid premature restrictions on open-weight models and sweeping distillation bans—Huang’s first X post. OpenAI, Anthropic, and Google did not sign. Lands as the White House weighs Chinese open-model limits.

Microsoft launched Frontier Company—$2.5B and ~6,000 embedded experts under Rodrigo Kede Lima—to co-design and deploy AI inside customer environments. Competing on deployment and outcomes, not only model access; customers retain co-built IP by default framing.

Microsoft made Copilot Cowork generally available worldwide for Microsoft 365 Copilot customers after Fortune 500-heavy preview. GA is when agent features stop being demos and start being credits, governance, and support SLAs.

Microsoft launched seven in-house MAI models at Build—flagship MAI-Thinking-1 (~35B active MoE, 256K) with vendor claims of blind preference over Sonnet 4.6 and SWE-Bench Pro parity with Opus 4.6. Cost-efficient Copilot/Foundry alternative while Azure stays multi-model.

NVIDIA and Microsoft position RTX Spark Windows PCs as the first client machines built for personal AI agents — 1 petaflop, up to 128GB unified memory, full-stack NVIDIA AI/graphics. The trade is latency-private local agents vs thin UIs over remote APIs; vendor power-efficiency claims are unaudited.

EY and Microsoft put >$1B over five years behind moving clients from AI pilots to enterprise-wide transformation on Azure, Copilot, Foundry, Fabric, and security. The trade is industrialized delivery vs experimentation — and another Big Four firm locking to a hyperscaler ecosystem.

GitHub makes GPT-5.3-Codex the base model for all Copilot Business and Enterprise orgs, replacing GPT-4.1 — OpenAI’s first LTS model, guaranteed through February 4, 2027. The trade is enterprise predictability (review cycles, security) over perpetual latest-model churn.

Microsoft Security’s MDASH — 100+ specialized agents in a multi-model scanning harness — helped find 16 Windows vulnerabilities (4 critical RCEs) shipping in the May 12 Patch Tuesday wave, while topping CyberGym-style benches in limited private preview. Dual-use agent stacks now sit next to the patch calendar.

Microsoft’s Q1 2026 Global AI Diffusion Report: 17.8% of working-age population used generative AI (+1.5 pp); UAE leads at 70.1%; US 31.3% (21st); 26 economies >30%. Git pushes +78% YoY; US developer employment at record ~2.2M. Microsoft telemetry adjusted for device/OS share — broadest single-vendor diffusion map, not a UN census.

Commerce’s CAISI announces agreements with Microsoft, Google DeepMind, and xAI for early frontier-model access and national-security evaluations — expanding prior OpenAI/Anthropic deals. Guardrails often stripped for probing; 40+ evals completed. Institutionalized voluntary vetting, not a binding pre-release statute.

At The Briefing: Financial Services, Anthropic launched ten Opus 4.7 prebuilt agents (pitchbook, KYC, GL recon, etc.) as Cowork/Code plugins or Managed Agents cookbooks, plus generally available Microsoft 365 Excel/PowerPoint/Word integrations with cross-app context and MCP connectors (Moody’s and peers). Packaging for regulated workflows — demo outcomes are customer showcases, not audited ROI.

Microsoft makes Agent 365 generally available ($15/user/mo standalone) and ships Microsoft 365 E7 Frontier Suite at $99/user/mo — bundling E5, Copilot, Entra Suite, and Agent 365 (~$18 under à-la-carte ~$117). First hyperscaler purpose-built, priced governance layer for agents — not bolted-on security chrome.

Microsoft FY26 Q3: $82.9B revenue (+18%), AI annual run rate $37B (+123%), Microsoft Cloud $54.5B (+29%), commercial RPO $627B (+99%). The trade is AI monetization at scale with Azure +40% — vs still-opaque AI margin mix and OpenAI-related non-GAAP noise.

Microsoft and OpenAI’s April 27 amendment keeps Azure first-ship primacy but lets OpenAI serve any cloud; IP license to Microsoft stays through 2032 but turns non-exclusive; OpenAI→Microsoft revenue share continues through 2030 with a total cap; Microsoft stops paying revenue share to OpenAI.

Microsoft Copilot Studio (March 2026 updates) adds native multi-agent orchestration: routing, state, shared tools for durable multi-step enterprise workflows — finance close, contract review, procurement cited. Governance/audit via enterprise identity. Platform bet on orchestrated specialists; early-customer claims are Microsoft-framed.