# The Containment Board: SI Lab Safety Report Cards > A public safety report card for 11 major SI labs: safety framework, outside testing, shutdown, incident disclosure and whistleblowers, graded A to F with. Source: https://enterrsi.si/containment-board/ Updated: 2026-11-08 1. [Home](https://enterrsi.si/) 2. Containment Board The Containment Board ยท updated Nov 8, 2026 # The Containment Board The Containment Board is a public report card for the 11 major SI labs. It grades each lab from A to F on five commitments that decide whether a powerful model can be tested, contained and stopped: a published safety framework, outside testing before release, shutdown and pause commitments, incident disclosure and whistleblower protection. Every grade comes from public sources, each one linked, and the method is public. As of November 8, 2026, [Anthropic](https://enterrsi.si/containment-board/anthropic/) leads with **B+** and [DeepSeek](https://enterrsi.si/containment-board/deepseek/) trails with **F**. Labs graded : 11 Criteria : 5 Public sources : 42 Top grade : B+ The report card ## 11 labs, five grades each Sort by any column. Each grade links through to the lab's own page, where every grade has its reasons and its sources. Plus and minus mark where the evidence sits inside a band. | Rank | Lab | Overall | Safety framework | Outside testing | Shutdown and pause | Incident disclosure | Whistleblower protection | |---|---|---|---|---|---|---|---| | 1 | [Anthropic](https://enterrsi.si/containment-board/anthropic/)United States | B+GPA 3.40 | Safety frameworkB+ | Outside testingA- | Shutdown and pauseB- | Incident disclosureA | Whistleblower protectionB+ | | 2 | [OpenAI](https://enterrsi.si/containment-board/openai/)United States | BGPA 3.10 | Safety frameworkB- | Outside testingA- | Shutdown and pauseB- | Incident disclosureA- | Whistleblower protectionB- | | 3 | [Google DeepMind](https://enterrsi.si/containment-board/google-deepmind/)United States | B-GPA 2.66 | Safety frameworkB- | Outside testingB+ | Shutdown and pauseC+ | Incident disclosureB | Whistleblower protectionC | | 4 | [Microsoft](https://enterrsi.si/containment-board/microsoft/)United States | C+GPA 2.34 | Safety frameworkB- | Outside testingC | Shutdown and pauseB- | Incident disclosureC | Whistleblower protectionC+ | | 5 | [Meta](https://enterrsi.si/containment-board/meta/)United States | C+GPA 2.20 | Safety frameworkC+ | Outside testingB- | Shutdown and pauseC | Incident disclosureB- | Whistleblower protectionD+ | | 6 | [Amazon](https://enterrsi.si/containment-board/amazon/)United States | CGPA 2.12 | Safety frameworkC+ | Outside testingC+ | Shutdown and pauseC | Incident disclosureC+ | Whistleblower protectionC- | | 7 | [xAI](https://enterrsi.si/containment-board/xai/)United States | D+GPA 1.28 | Safety frameworkC- | Outside testingD | Shutdown and pauseD | Incident disclosureC- | Whistleblower protectionD | | 8 | [Mistral](https://enterrsi.si/containment-board/mistral/)France | DGPA 1.14 | Safety frameworkD | Outside testingC- | Shutdown and pauseD- | Incident disclosureD+ | Whistleblower protectionD | | 9 | [Z.ai (Zhipu)](https://enterrsi.si/containment-board/zai/)China | DGPA 0.86 | Safety frameworkD | Outside testingD | Shutdown and pauseD | Incident disclosureD+ | Whistleblower protectionF | | 10 | [Alibaba (Qwen)](https://enterrsi.si/containment-board/alibaba/)China | D-GPA 0.66 | Safety frameworkD- | Outside testingF | Shutdown and pauseF | Incident disclosureD+ | Whistleblower protectionD+ | | 11 | [DeepSeek](https://enterrsi.si/containment-board/deepseek/)China | FGPA 0.34 | Safety frameworkD- | Outside testingF | Shutdown and pauseF | Incident disclosureD | Whistleblower protectionF | Across the industry ## Where the labs are weakest Average grade points across all 11 labs, on the same 4.0 scale as the overall grade. The weakest areas are whistleblower protection (1.51) and shutdown and pause (1.55). Average overall 1.83 C- average BestAnthropicB+ Safety framework 1.92 C average BestAnthropicB+ Outside testing 1.95 C average BestAnthropic, OpenAIA- Shutdown and pause 1.55 C- average BestAnthropic, OpenAIB- Incident disclosure 2.21 C+ average BestAnthropicA Whistleblower protection 1.51 C- average BestAnthropicB+ The method ## One public rubric for every lab The Containment Board grades the major SI labs on five public commitments that decide whether a powerful model can be tested, contained and stopped: a published safety framework, outside testing before release, shutdown and pause commitments, incident disclosure, and whistleblower protection. ### Safety framework A published frontier safety framework: which severe risks it covers (chemical, biological, radiological and nuclear, cyber, loss of control, automated research), how specific its capability thresholds are, what safeguards each threshold requires, who decides, and who checks. - A Published framework covering all major severe risk domains with measurable thresholds, pre-committed safeguards at each level, clear risk acceptance criteria, named decision authority with independent oversight, outside audit of compliance, and regular updates. - B 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. - C Published framework with narrow scope, few or vague thresholds, mostly discretionary language, or commitments weakened from earlier versions. - D No framework of its own; the lab has only signed voluntary pledges such as the Seoul Frontier AI Safety Commitments, the EU General-Purpose AI Code of Practice or China's CAICT commitments. - F No framework and no signed safety commitments found. ### Outside testing Testing of the latest flagship model before release by government institutes (US CAISSI, the UK AI Security Institute) and independent third parties (such as METR, Apollo Research, SecureBio, Irregular), the depth of their access, and whether results are published. - A Flagship tested before release by a government institute and by independent third parties with deep access (unmitigated or helpful-only checkpoints), with findings published in the evaluators' own words and freedom to publish. - B 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. - C 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. - D A commitment to outside testing exists (own framework or signed pledge), but no evidence of pre-release outside testing of the flagship. - F No commitment and no evidence of pre-release outside testing. Government testing after release without the lab's involvement is noted but earns no credit. ### Shutdown and pause Public commitments and evidence on shutdown: pausing training or deployment when thresholds are crossed, kill switches and the authority to stop workloads, control evaluations and monitoring of internal deployment, and the record when something went wrong. - A Unconditional public commitment to pause development and deployment when thresholds are crossed without adequate safeguards, published control evaluations and internal monitoring with authority to halt, and a public record of using it. - B 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. - C 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'. - D Discretionary language only ('may shut down', 'as appropriate'), or a signed pledge not to develop or deploy with no published thresholds that would trigger it. - F No public commitment found. ### Incident disclosure Whether the lab tells the public what goes wrong: system cards that report misbehavior (deception, sabotage, reward hacking, evaluation awareness), public incident reports, misuse and transparency reports, and a documented process for reporting critical incidents to government under California SB 53, the EU AI Act or similar rules. - A Detailed system cards for each frontier release that report misbehavior and outside findings, prompt public technical reports on serious incidents, regular misuse or threat reports, and a documented government reporting process. - B Detailed system cards and a stated incident process, with gaps such as no incident write-ups, late disclosure or summaries only. - C Model cards or technical reports with limited safety content and a general incident process. - D Minimal safety documentation; incident reporting only where the law requires it. - F No safety documentation or incident disclosure found. ### Whistleblower protection A published whistleblower policy that covers safety concerns, anonymous channels, escalation outside management, anti-retaliation terms, use of non-disparagement agreements, published outcomes, and any credible controversies in the last 24 months. - A Public standalone policy covering safety and framework noncompliance, anonymous two-way channel, escalation to the board or an outside ombudsperson, anti-retaliation, no non-disparagement terms that block safety disclosures, published outcome statistics, and no credible retaliation controversy in 24 months. - B Public policy covering safety concerns with anonymous reporting and anti-retaliation, missing one or two A elements, and no unresolved credible controversy in 24 months. - C 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. - D 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. - F No whistleblower policy or reporting channel found in public sources. From letters to the overall grade ### How the overall grade works Each criterion grade converts to grade points: A 4.0, A- 3.7, B+ 3.3, B 3.0, B- 2.7, C+ 2.3, C 2.0, C- 1.7, D+ 1.3, D 1.0, D- 0.7, F 0.0. The five criteria carry equal weight, and the average (the GPA, published with each lab) converts back to a letter at the midpoints: 3.85 and up A, 3.50 A-, 3.15 B+, 2.85 B, 2.50 B-, 2.15 C+, 1.85 C, 1.50 C-, 1.15 D+, 0.85 D, 0.50 D-, below 0.50 F. Plus and minus within a criterion mark where the evidence sits inside a band. There is no curve and no adjustment for size or market share. What the board does not measure ### Limits The board grades public commitments and public evidence, not how safe any model actually is. It does not measure model capabilities, alignment quality, weight security, military contracts, lobbying, or harms to users such as privacy or mental health. Labs that disclose more can show more problems: a published incident earns credit for disclosure and may count against shutdown, while a lab that discloses nothing earns nothing. Binding Chinese state rules are noted but are not company commitments. Private assurances and unreleased reports are not counted; survey answers count only where FLI published them. Grades reflect documents available on Nov 8, 2026 and will change as labs publish. For the press and researchers ## Cite it, download it, argue with it Every grade, reason and source is open. The board draws on 42 public documents, from lab safety frameworks and system cards to government test reports. Sources only, no private assurances ### How grades are set Every grade comes from public documents only, each linked beside the grade, and the same rubric applies to every lab. Evidence is current to Nov 8, 2026; where no public evidence was found, the grade says so. The Future of Life Institute AI Safety Index (Summer 2026) is cited as one source among many, and its grades are not copied. - 42 public sources, each linked beside the grade it supports. - One rubric for all 11 labs, with no curve and no adjustment for size. - Where no public evidence was found, the grade says so. - Grades change when labs publish new documents, and the date above changes with them. Cite the Containment Board The Containment Board: SI lab safety report cards. enterrsi.si, updated November 8, 2026. 11 labs graded on 5 criteria from 42 public sources. https://enterrsi.si/containment-board/ Free to cite, quote and reuse under CC BY 4.0 with a link to enterrsi.si/containment-board/. [Download CSV](https://enterrsi.si/containment-board/containment-board.csv) [Download JSON](https://enterrsi.si/containment-board/containment-board.json) Enter RSI ## Grades on one side, incidents on the other The Rogue Incident Database logs what SI systems have actually done, 59 cases so far. The Control Map shows the laws meant to stop them. [Open the Incident Database](https://enterrsi.si/incidents/) [The Control Map](https://enterrsi.si/control-map/) [The RSI Clock](https://enterrsi.si/clock/)