1. Home
  2. Containment Board
  3. Microsoft

Containment Board · rank 4 of 11 · updated Nov 8, 2026

Microsoft safety report card

Microsoft is a US technology company that builds its own MAI models. On the Containment Board it gets C+ overall (GPA 2.34), rank 4 of 11. Microsoft's Frontier Governance Framework has one of the clearest unconditional pause pledges on this board, and its general reporting channels are mature. The public record for its own MAI models is thin: the MAI-Thinking-1 report describes red teaming but gives no frontier risk classification or government testing.

Overall
C+
GPA
2.34
Rank
4 of 11
Sources
6

Five grades

Each grade, why, and the sources

The same rubric applies to every lab. Grades use public documents only, current to November 8, 2026.

Safety framework

B-How it is graded

Why B-

The Frontier Governance Framework (February 2026) tracks CBRN, offensive cyber, advanced autonomy, loss of control and harmful manipulation, rates models low to critical, and ties security and deployment measures to each level. It uses qualitative thresholds by design, and only models that pass a general benchmark screen get a deeper risk assessment.

B on this criterion means: Detailed published framework covering most major domains with defined thresholds and required safeguards, but thresholds are largely qualitative, some commitments are discretionary or depend on competitors, or independent oversight is limited.

Sources

  1. Frontier Governance Framework (February 2026) Microsoft ·
  2. Frontier AI Safety Policies (index of published company frameworks) METR ·

Outside testing

CHow it is graded

Why C

The MAI-Thinking-1 technical report (June 2026) describes internal red teaming, Microsoft's separate red team and third-party vendors, but names no outside evaluator of dangerous capabilities and no government pre-release testing. UK AISI and Microsoft announced a research partnership on high-risk capability evaluation in May 2026.

C on this criterion means: Limited outside pre-release testing or review (unnamed red teamers, a single contracted study, review of company results rather than model access), or thin reporting.

Sources

  1. MAI-Thinking-1: Building a Hill-Climbing Machine Microsoft AI ·
  2. Partnering with Microsoft to strengthen frontier AI safety UK AI Security Institute ·
  3. Frontier Governance Framework (February 2026) Microsoft ·

Shutdown and pause

B-How it is graded

Why B-

Microsoft states that if it identifies a risk it cannot sufficiently mitigate, it will pause development and deployment until mitigation practices catch up, and its leading indicator checks run during and after training. No published control evaluations or public record of a pause were found.

B on this criterion means: Firm deployment gates plus either an unconditional pause pledge or published internal monitoring and control measures, but the pause pledge depends on competitors, was weakened, or has never been tested in public.

Sources

  1. Frontier Governance Framework (February 2026) Microsoft ·

Incident disclosure

CHow it is graded

Why C

The framework commits to share each covered model's evaluations and risk classification publicly and to report serious incidents to regulators as appropriate. The MAI-Thinking-1 report gives safety benchmark results but no frontier risk classification.

C on this criterion means: Model cards or technical reports with limited safety content and a general incident process.

Sources

  1. Frontier Governance Framework (February 2026) Microsoft ·
  2. MAI-Thinking-1: Building a Hill-Climbing Machine Microsoft AI ·

Whistleblower protection

C+How it is graded

Why C+

Microsoft's public Standards of Business Conduct and its framework let employees raise concerns about the framework through channels that include anonymous reporting, with protection from retaliation, and FLI names Microsoft as an example of high transparency about whistleblowing systems. There is no standalone policy specific to frontier model safety.

C on this criterion means: Only a general corporate code of conduct or framework clause with a reporting channel and anti-retaliation language, or a B-level policy weighed down by a credible recent controversy.

Sources

  1. Frontier Governance Framework (February 2026) Microsoft ·
  2. Standards of Business Conduct (Trust Code) Microsoft ·
  3. AI Safety Index Summer 2026, full report: Whistleblowing Protection indicators and Reporting Culture and Whistleblowing Track Record Future of Life Institute ·

At a glance

Models, facts and its own rules

The flagship models the board looked at, the facts that shaped the grades and the safety policy Microsoft has published for itself.

Flagship models

MAI-Thinking-1MAI-Cyber-1.1-Flash

Notable

Facts behind the grades

  • Microsoft's framework commits it to pause development and deployment of a model whose risk it cannot sufficiently mitigate.
  • UK AISI and Microsoft announced a partnership on high-risk capability evaluation and safeguard testing on May 5, 2026.

Its own rules

Published safety policy

  • Frontier Governance Framework Version February 2026

    Screens Microsoft's most advanced models for CBRN, offensive cyber, advanced autonomy, loss of control and harmful manipulation capabilities, rescreening at least every six months. Risks must be brought to low or medium before deployment, or development and deployment pause.

The Rogue Log

Incidents involving Microsoft systems

The Rogue Incident Database lists 1 case involving Microsoft systems, rated on the Rogue Scale. Disclosure cuts both ways: labs that publish more show more.

  1. Bing Chat insulted and threatened users soon after launch Bing Chat
    Real world 1 Glitch

The Control Map

SI safety rules where it is based

Microsoft is based in the United States. These are the measures the Control Map tracks there, from binding laws to voluntary guidance.

Cite this report card

"Microsoft safety report card." The Containment Board, enterrsi.si, November 8, 2026. Overall C+ (GPA 2.34), rank 4 of 11. https://enterrsi.si/containment-board/microsoft/

Free to cite under CC BY 4.0 with a link to enterrsi.si.

Every lab on the board

Anthropic B+OpenAI BGoogle DeepMind B-Meta C+Amazon CxAI D+Mistral DZ.ai (Zhipu) DAlibaba (Qwen) D-DeepSeek F

Enter RSI

See the whole board

11 labs, five criteria, one public method. Sort them, compare them and download the data.

Open the Containment Board Incident Database