AI Governance Institute
← News
Corporate Policy2026-09-15

Microsoft's MAI Code of Conduct Sets Agentic Chain-of-Command Standard

What happened

Microsoft released a draft Humanist AI Code of Conduct for its MAI Models, establishing what the company calls Absolute Constraints. These constraints block models from generating working exploit code, attack tooling, intrusion procedures, and evasion techniques regardless of how requests are framed. The policy also introduces a formal Chain of Command authority model for agentic systems: sub-agents must operate with minimum privilege and are constrained to the same scope and permissions as the originating model. Visible reasoning is required throughout agent task execution, addressing a structural gap that researchers and security teams have flagged repeatedly in recent agentic incident reports. A six-week public consultation period precedes a revised version that will guide 2027 MAI Model development. Organizations with legitimate defensive cybersecurity or dual-use research needs will face an enhanced review track to access capabilities beyond standard settings.

Why it matters

  • ·The Chain of Command authority model sets an explicit vendor standard for minimum-privilege agentic operation. Enterprises running MAI Models in multi-agent pipelines must now verify their own deployments mirror these scope and permission constraints, or risk a documented divergence from the vendor's own governance baseline.
  • ·The Absolute Constraints on offensive cyber outputs create a reference point for dual-use intake reviews. Organizations in defensive security, penetration testing, or research functions must establish a documented process for requesting enhanced capability access under the new enhanced review track, or those use cases become ungoverned.
  • ·The six-week consultation window and 2027 development roadmap signal give compliance teams a narrow, time-bound opportunity to submit feedback that shapes binding model behavior. Organizations that miss this window lose influence over constraints that will govern models they may already be deploying.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

AGT-001Agent Permission BoundariesAGT-004Multi-Agent Trust HierarchyAGT-007Agent Scope and Task BoundariesSCT-005National Security and Dual-Use AI Risk AssessmentPRC-006Vendor Safety Commitment Verification

What to do now

  • ☐Review all agentic deployments using MAI Models against the Chain of Command requirements: confirm sub-agents operate under minimum privilege and inherit, not exceed, the originating model's permission scope.
  • ☐Identify any internal use cases involving defensive cybersecurity or dual-use research that may require capabilities beyond standard MAI Model settings, and open a formal intake process for the enhanced review track before the consultation closes.
  • ☐Submit organizational feedback during the six-week consultation period, particularly on the enhanced review criteria, to shape 2027 model constraints that will affect existing and planned deployments.
  • ☐Update your vendor governance change monitoring process to track the revised code once it is published post-consultation, and trigger a re-assessment of any affected agentic deployment documentation.
  • ☐Map the Absolute Constraints against your existing dual-use AI risk assessment to identify any current deployments that may be requesting model outputs that fall near or within the prohibited categories.

What to watch next

Compliance teams should monitor the outcome of the six-week consultation for constraint changes that could affect defensive security and dual-use research use cases. The revised code, expected before the 2027 development cycle, will likely become a reference standard for enterprise vendor benchmarking and could influence regulatory expectations around agentic minimum-privilege requirements. Watch for whether competing frontier labs issue comparable authority models in response, as a convergent industry baseline would raise the bar for what regulators and auditors expect from enterprise deployers. The parallel trajectory of agent governance becoming binding across multiple jurisdictions makes this vendor-level code an early signal of where formal requirements may land.

Related Coverage

Corporate Policy2026-11-09

Microsoft Publishes Agent Risk Controls That Shift Authorization Outside the Model

Microsoft has published guidance titled "Reduce autonomous agentic AI risk" setting out how enterprises should secure AI agents across tool, service, and agent-to-agent interactions. The core requirement is that filtering and authorization controls must sit outside the AI model itself, not inside it. The guidance treats all content retrieved by agents as untrusted by default.

Research2026-11-13

EchoLeak Proves AI Assistants Cannot Defend Against Prompt Injection Alone

A flaw named EchoLeak was patched by Microsoft in May 2025. It showed that a crafted email could cause Microsoft Copilot to silently send sensitive data to an attacker through an automatically loaded image. Security researchers publishing in November 2026 argue the underlying weakness persists across all AI assistants, because no AI model can reliably distinguish legitimate content from hidden attacker instructions. Enterprises using AI assistants with access to email, documents, or internal systems need architecture-level controls that sit outside the model itself.

Corporate Policy2026-11-11

IBM and Microsoft's Imagine Enterprise Puts Agentic Governance Claims to the Test

IBM Consulting and Microsoft have jointly launched Imagine Enterprise, an agentic AI platform targeting regulated industries including financial services, insurance, and retail. The platform combines IBM Consulting's library of over 6,000 prebuilt AI agents with Microsoft's Azure cloud, Entra identity management, Purview data governance, and Defender and Sentinel security tooling. Enterprises using the platform remain responsible for satisfying their own regulatory obligations, even where the platform supplies governance tooling.