Daily Agentic Field Watch - 2026-10-01 21:00 UTC

• Microsoft 365 G7 and Agent 365 became purchasable for U.S. Government Community Cloud customers today, bringing Copilot/agent capabilities and the Agent 365 governance control plane into that regulated market. Microsoft says capabilities will phase in as workloads satisfy GCC authorization milestones, so this is a licensing/availability milestone rather than evidence that every workload is live on day one. Source: https://www.microsoft.com/en-us/microsoft-cloud/blog/industry-government/government-operations-and-infrastructure/2026/09/15/introducing-microsoft-365-g7-intelligence-trust-for-the-mission-ahead/
• MongoDB launched Atlas Agent Engine in public preview, bundling agent runtime/execution, persistent memory/context, state management, retrieval and governance/guardrails as a model- and framework-flexible production layer. Public preview is currently on AWS with consumption-based pricing for runtime and memory; this is a notable infrastructure entrant addressing production plumbing, though vendor claims of solving the “production bottleneck” have not yet been independently validated. Sources: https://www.mongodb.com/company/newsroom/press-releases/mongodb-launches-atlas-agent-engine-to-put-ai-agents-in-production-without-a-new-stack ; https://www.mongodb.com/products/updates/ ; https://www.reddit.com/r/MongoDB_Official/comments/1wtccqf/just_announced_mongodb_90_atlas_infinite_and/
• Microsoft Defender for Office 365 Plan 2 now detects prompt-injection instructions in inbound email—including hidden/obfuscated content—before delivery to users or AI assistants, as part of mail-flow inspection without additional configuration. This is a concrete defense at the ingestion layer for email-grounded copilots and agents; Defender runtime safety and XDR correlation remain additional layers, not guarantees that all prompt injection is eliminated. Sources: https://learn.microsoft.com/en-us/defender-office-365/step-by-step-guides/prompt-injection-protection-defender-for-office-365 ; https://techcommunity.microsoft.com/blog/MicrosoftDefenderforOffice365Blog/defending-the-inbox-against-prompt-injection-attacks/4534636
• Emerging-model radar / availability change: OpenAI’s GPT-5.4-Cyber API model was removed today (Oct. 1), so agents or security-analysis pipelines pinned to that model ID need a replacement; the official deprecation notice recommends migrating to the most capable cyber model available to the account but does not name a universal successor. Source: https://developers.openai.com/api/docs/deprecations