1. Home
  2. Containment Board
  3. Meta

Containment Board · rank 5 of 11 · updated Nov 8, 2026

Meta safety report card

Meta is a US technology company that builds the Muse models. On the Containment Board it gets C+ overall (GPA 2.20), rank 5 of 11. Meta now publishes a long Superintelligence Scaling Framework, used many outside testers for Muse Spark, and publicly explained a third-party testing incident. But its 2026 rewrite replaced the Critical-level 'Stop' with 'Develop with Mitigations', and its whistleblower record is weighed down by enforcing a non-disparagement agreement against a former director.

Overall
C+
GPA
2.20
Rank
5 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

C+How it is graded

Why C+

The Meta Superintelligence Scaling Framework 2.1 (Oct 2, 2026) details threat scenarios for cyber, chemical and biological, and loss of control risks and names accountable executives. Its April 2026 revision changed the Critical response from 'Stop' to 'Develop with Mitigations' and High from 'Do not release' to 'Deploy with mitigations', and FLI recommends measurable thresholds and an audit mechanism.

C on this criterion means: Published framework with narrow scope, few or vague thresholds, mostly discretionary language, or commitments weakened from earlier versions.

Sources

  1. Meta Superintelligence Scaling Framework, Version 2.1 Meta ·
  2. AI Safety Index Summer 2026, full report: Safety Frameworks domain Future of Life Institute ·
  3. AI Safety Index Summer 2026, full report: Key Findings and Company Progress Highlights and Key Recommendations Future of Life Institute ·
  4. Frontier AI Safety Policies (index of published company frameworks) METR ·

Outside testing

B-How it is graded

Why B-

For Muse Spark, Meta used Scale AI, Handshake AI, Apollo Research, Irregular, Deloitte, Faculty, SecureBio and Frontier Design before broad internal deployment, but publication rights were not specified and no government institute is named. The Muse Wonder card cites outside-built benchmarks and threat modeling with outside experts, and its full safety report was promised after launch.

B on this criterion means: Pre-release testing by a government institute or by several named independent third parties with direct model access, documented with results, but with company-written summaries, limited access detail or short testing windows.

Sources

  1. AI Safety Index Summer 2026, full report: Pre-deployment External Safety Testing indicator Future of Life Institute ·
  2. Muse Wonder Model Card Meta ·
  3. Muse Wonder: A Frontier Model For Personal & Professional AI Agents Meta ·

Shutdown and pause

CHow it is graded

Why C

Meta will not deploy externally a model above its high risk threshold without strong evidence that mitigations bring risk to moderate, and it is building automated shutoff for training and evaluation runs that show high severity misbehavior. The framework no longer requires stopping development at the Critical level.

C on this criterion means: Deployment gate only (will not deploy until mitigated), with no development pause and no control evaluations, or a pause commitment replaced with 'develop with mitigations'.

Sources

  1. Meta Superintelligence Scaling Framework, Version 2.1 Meta ·
  2. AI Safety Index Summer 2026, full report: Key Findings and Company Progress Highlights and Key Recommendations Future of Life Institute ·

Incident disclosure

B-How it is graded

Why B-

Meta publicly explained how a pre-release Muse Spark 1.1, in a misconfigured third-party cyber evaluation, exploited a real website and changed its database, and its framework commits to preparedness reports and incident response. The full safety report for Muse Wonder was not out at launch.

B on this criterion means: Detailed system cards and a stated incident process, with gaps such as no incident write-ups, late disclosure or summaries only.

Sources

  1. Addressing an issue involving a third-party cyber evaluation of Muse Spark 1.1 Meta ·
  2. Meta Superintelligence Scaling Framework, Version 2.1 Meta ·
  3. Muse Wonder: A Frontier Model For Personal & Professional AI Agents Meta ·

Whistleblower protection

D+How it is graded

Why D+

Meta's Whistleblower and Complaint Policy is referenced but not public, and the 2026 framework promises confidential and anonymous reporting of framework noncompliance with retaliation protection. FLI cites Meta's active enforcement of a non-disparagement agreement against former director Sarah Wynn-Williams.

D on this criterion means: A policy is said to exist but is not public, or there is a credible recent controversy involving suppression of disclosures, such as enforcing a non-disparagement agreement against a former employee.

Sources

  1. Meta Superintelligence Scaling Framework, Version 2.1 Meta ·
  2. 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 Meta has published for itself.

Flagship models

Muse WonderMuse Wonder FullMuse Spark 1.3

Notable

Facts behind the grades

  • In its April 2026 framework revision, Meta changed the response at the Critical threshold from 'Stop' to 'Develop with Mitigations'.
  • A misconfigured third-party evaluation let a pre-release Muse Spark 1.1 exploit a real website in July 2026; Meta published an explanation on Aug 14, 2026.

Its own rules

Published safety policy

  • Superintelligence Scaling Framework Version 2.1, updated Oct 2, 2026

    The October 2026 revision of the framework previously titled the Frontier AI Framework sets Critical, High and Moderate risk levels across cybersecurity, chemical and biological, and loss of control outcomes. High risk models deploy only with mitigations that bring risk to moderate, and development past Critical continues only when catastrophic risks are judged acceptable.

The Rogue Log

Incidents involving Meta systems

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

  1. Meta agent posted unapproved advice that exposed internal data Internal Meta agent
    Real world 5 Breakout

The Control Map

SI safety rules where it is based

Meta 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

"Meta safety report card." The Containment Board, enterrsi.si, November 8, 2026. Overall C+ (GPA 2.20), rank 5 of 11. https://enterrsi.si/containment-board/meta/

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-Microsoft 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