# The Promise Tracker: SI Lab Safety Promises, Checked

> 47 public safety promises by 11 SI labs and their leaders, each with its primary source and a status: kept, not met, withdrawn or in progress.

Source: https://enterrsi.si/promises/
Updated: 2026-11-08

1. [Home](https://enterrsi.si/)
2. Promise Tracker

The Promise Tracker · checked Nov 10, 2026

# The Promise Tracker

The Promise Tracker lists **47 public safety promises** made by the 11 SI labs on the [Containment Board](https://enterrsi.si/containment-board/) and their leaders, each with its primary source, its date and a status. A promise is marked Kept or Not met only when a reputable public source reports it, and that source is named and linked. As of November 10, 2026, 11 are reported kept, 4 not met and 2 withdrawn; the other 30 are in progress or have no public evidence yet.

Promises
:   47

Labs
:   11

Reported kept
:   11

Reported not met
:   4

The tally

## What the labs promised, and what happened

Every status comes with a named public source, except the two cautious ones. The bars show each lab's promises by status; the chip is its Containment Board grade.

47

Promises

11

Kept

4

Not met

2

Withdrawn

18

In progress

12

No evidence yet

1. [Anthropic](https://enterrsi.si/promises/anthropic/)5B+
2. [OpenAI](https://enterrsi.si/promises/openai/)7B
3. [Google DeepMind](https://enterrsi.si/promises/google-deepmind/)5B-
4. [Microsoft](https://enterrsi.si/promises/microsoft/)6C+
5. [Meta](https://enterrsi.si/promises/meta/)5C+
6. [Amazon](https://enterrsi.si/promises/amazon/)5C
7. [xAI](https://enterrsi.si/promises/xai/)4D+
8. [Mistral](https://enterrsi.si/promises/mistral/)3D
9. [Z.ai (Zhipu)](https://enterrsi.si/promises/zai/)3D
10. [Alibaba (Qwen)](https://enterrsi.si/promises/alibaba/)2D-
11. [DeepSeek](https://enterrsi.si/promises/deepseek/)2F

Every promise

## All 47 promises, newest first

Filter by status, lab or topic. Each entry gives the exact words when they are short enough, the primary source, the status with its own source and the day it was checked.

No public evidence yet[Mistral](https://enterrsi.si/promises/mistral/)Oct 6, 2026

### Red-team the model with cybersecurity leaders, partners and state authorities before releasing its weights.

Mistral AI, announcing Mistral Large 4

Exact words

> "we are red-teaming the model in real-world settings with cybersecurity leaders, vetted partners, and state authorities"

Source [Introducing Mistral Large 4](https://mistral.ai/news/mistral-large-4/) Mistral AI · Oct 6, 2026

**No public evidence yet.** No evaluator, method or result has been published; the weights were released on Oct 29, 2026.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Meta](https://enterrsi.si/promises/meta/)Oct 2, 2026

### Not release a high-risk model externally without strong evidence that mitigations bring the risk down.

Meta, Superintelligence Scaling Framework 2.1

Exact words

> "we will not deploy the model externally unless we have strong additional evidence"

Source [Meta Superintelligence Scaling Framework, Version 2.1](https://research.meta.ai/static/meta-superintelligence-scaling-framework) Meta · Oct 2, 2026

**In progress.** The framework took effect in October 2026; Meta has not published an outside audit of it.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Amazon](https://enterrsi.si/promises/amazon/)Sep 17, 2026

### Not deploy a model that meets a Critical Capability Threshold until safeguards mitigate the risk.

Amazon, Frontier Model Safety Framework

Exact words

> "we will not deploy the model until safeguards appropriately mitigate the risks."

Source [Amazon's Frontier Model Safety Framework (September 2026)](https://cdn.amazon.science/a0/4f/bca5cb32495280835d9370a466b7/amazon-fmsf-9-2026.pdf) Amazon · Sep 17, 2026

**In progress.** METR concluded that Nova 2.0 Lite does not cross the automated research Critical Capability Threshold.

- [Strengthening AI safety evaluation through independent third-party review](https://www.amazon.science/blog/strengthening-ai-safety-evaluation-through-independent-third-party-review) Amazon Science · Oct 20, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Amazon](https://enterrsi.si/promises/amazon/)Sep 17, 2026

### Publish safety evaluation reports for its most capable frontier models.

Amazon, Frontier Model Safety Framework

Exact words

> "we publish safety evaluation reports for our most capable frontier models"

Source [Amazon's Frontier Model Safety Framework (September 2026)](https://cdn.amazon.science/a0/4f/bca5cb32495280835d9370a466b7/amazon-fmsf-9-2026.pdf) Amazon · Sep 17, 2026

**In progress.** An evaluation report for Nova 2.0 Lite was posted to arXiv by Amazon authors on Jan 27, 2026.

- [Evaluating Nova 2.0 Lite model under Amazon's Frontier Model Safety Framework](https://arxiv.org/abs/2601.19134) arXiv (Amazon authors) · Jan 27, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Anthropic](https://enterrsi.si/promises/anthropic/)Jul 8, 2026

### Publish Risk Reports on its deployed models, and let outside reviewers check them when the Long-Term Benefit Trust asks.

Anthropic, Responsible Scaling Policy 3.4

Summary in our words

Source [Anthropic's Responsible Scaling Policy](https://www.anthropic.com/responsible-scaling-policy) Anthropic · Jul 8, 2026

**In progress.** Version 3.4 took effect on Jul 8, 2026; no outside review of a Risk Report has been published yet.

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[xAI](https://enterrsi.si/promises/xai/)Jun 30, 2026

### Send authorities a copy of incident reports.

xAI, Frontier Artificial Intelligence Framework

Exact words

> "will provide the relevant authorities with a copy of the incident report"

Source [xAI Frontier Artificial Intelligence Framework](https://media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-f1f0c916.pdf) xAI · Jun 30, 2026

**No public evidence yet.** FLI found no misuse or misalignment transparency reports from xAI (Jul 14, 2026).

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Google DeepMind](https://enterrsi.si/promises/google-deepmind/)Apr 17, 2026

### Back any model that reaches a Critical Capability Level with a safety case before deployment.

Google DeepMind, Frontier Safety Framework 3.1

Exact words

> "Where the model has reached a CCL, the residual risk assessment will be supplemented with a safety case."

Source [Frontier Safety Framework, Version 3.1](https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf) Google DeepMind · Apr 17, 2026

**In progress.** Google's Gemini 4 Argon safety report (Oct 13, 2026) says it cannot rule out the cybersecurity Critical Capability Level; no safety case has been published.

- [Gemini 4 Argon Safety and Security Report](https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf) Google DeepMind · Oct 13, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Microsoft](https://enterrsi.si/promises/microsoft/)Feb 13, 2026

### Pause development and deployment of a model whose risk it cannot sufficiently mitigate.

Microsoft, Frontier Governance Framework

Exact words

> "we will pause development and deployment until the point at which mitigation practices evolve to meet the risk"

Source [Frontier Governance Framework (February 2026)](https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/Frontier-Governance-Framework-Feb-2026.pdf) Microsoft · Feb 13, 2026

**In progress.** No public record of a pause has been reported.

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[Microsoft](https://enterrsi.si/promises/microsoft/)Feb 13, 2026

### Publish each covered model's evaluations and its risk classification.

Microsoft, Frontier Governance Framework

Exact words

> "relevant evaluations, and the model's risk classification will be shared publicly"

Source [Frontier Governance Framework (February 2026)](https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/Frontier-Governance-Framework-Feb-2026.pdf) Microsoft · Feb 13, 2026

**No public evidence yet.** The MAI-Thinking-1 technical report (Jun 2, 2026) gives safety benchmark results but no frontier risk classification.

- [MAI-Thinking-1: Building a Hill-Climbing Machine](https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf) Microsoft AI · Jun 2, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Microsoft](https://enterrsi.si/promises/microsoft/)Nov 6, 2025

### Build superintelligence only within limits that keep humans in control.

Mustafa Suleyman, CEO of Microsoft AI

Exact words

> "we are not building a superintelligence at any cost, with no limits"

Source [Towards Humanist Superintelligence](https://microsoft.ai/blog/towards-humanist-superintelligence) Microsoft AI · Nov 6, 2025

**In progress.** Microsoft's Frontier Governance Framework (Feb 2026) puts limits in writing, including a pause commitment.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Amazon](https://enterrsi.si/promises/amazon/)Aug 1, 2025

### Follow the Safety and Security chapter of the EU General-Purpose AI Code of Practice for its most capable models.

Amazon, a signatory of the EU Code of Practice

Summary in our words

Source [The General-Purpose AI Code of Practice: signatories](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) European Commission · Oct 7, 2026

**In progress.** The European Commission lists the signature (page updated Oct 7, 2026); the EU AI Office has not published an assessment of compliance.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Anthropic](https://enterrsi.si/promises/anthropic/)Aug 1, 2025

### Follow the Safety and Security chapter of the EU General-Purpose AI Code of Practice for its most capable models.

Anthropic, a signatory of the EU Code of Practice

Summary in our words

Source [The General-Purpose AI Code of Practice: signatories](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) European Commission · Oct 7, 2026

**In progress.** The European Commission lists the signature (page updated Oct 7, 2026); the EU AI Office has not published an assessment of compliance.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Google DeepMind](https://enterrsi.si/promises/google-deepmind/)Aug 1, 2025

### Follow the Safety and Security chapter of the EU General-Purpose AI Code of Practice for its most capable models.

Google, a signatory of the EU Code of Practice

Summary in our words

Source [The General-Purpose AI Code of Practice: signatories](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) European Commission · Oct 7, 2026

**In progress.** The European Commission lists the signature (page updated Oct 7, 2026); the EU AI Office has not published an assessment of compliance.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Microsoft](https://enterrsi.si/promises/microsoft/)Aug 1, 2025

### Follow the Safety and Security chapter of the EU General-Purpose AI Code of Practice for its most capable models.

Microsoft, a signatory of the EU Code of Practice

Summary in our words

Source [The General-Purpose AI Code of Practice: signatories](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) European Commission · Oct 7, 2026

**In progress.** The European Commission lists the signature (page updated Oct 7, 2026); the EU AI Office has not published an assessment of compliance.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Mistral](https://enterrsi.si/promises/mistral/)Aug 1, 2025

### Follow the Safety and Security chapter of the EU General-Purpose AI Code of Practice for its most capable models.

Mistral AI, a signatory of the EU Code of Practice

Summary in our words

Source [The General-Purpose AI Code of Practice: signatories](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) European Commission · Oct 7, 2026

**In progress.** The European Commission lists the signature (page updated Oct 7, 2026); the EU AI Office has not published an assessment of compliance.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[OpenAI](https://enterrsi.si/promises/openai/)Aug 1, 2025

### Follow the Safety and Security chapter of the EU General-Purpose AI Code of Practice for its most capable models.

OpenAI, a signatory of the EU Code of Practice

Summary in our words

Source [The General-Purpose AI Code of Practice: signatories](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) European Commission · Oct 7, 2026

**In progress.** The European Commission lists the signature (page updated Oct 7, 2026); the EU AI Office has not published an assessment of compliance.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[xAI](https://enterrsi.si/promises/xai/)Aug 1, 2025

### Follow the Safety and Security chapter of the EU General-Purpose AI Code of Practice for its most capable models.

xAI, a signatory of the EU Code of Practice

Summary in our words

Source [The General-Purpose AI Code of Practice: signatories](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) European Commission · Oct 7, 2026

**In progress.** The European Commission lists the signature (page updated Oct 7, 2026); the EU AI Office has not published an assessment of compliance. xAI signed the Safety and Security chapter only.

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[Meta](https://enterrsi.si/promises/meta/)Jul 30, 2025

### Be rigorous about superintelligence risks and careful about what Meta open sources.

Mark Zuckerberg, CEO of Meta

Exact words

> "We'll need to be rigorous about mitigating these risks and careful about what we choose to open source."

Source [Personal Superintelligence](https://www.meta.com/superintelligence/) Meta · Jul 30, 2025

**No public evidence yet.** No public, reputable assessment of this pledge was found.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[OpenAI](https://enterrsi.si/promises/openai/)Apr 15, 2025

### Stop developing a model that reaches a Critical capability until matching safeguards exist.

OpenAI, Preparedness Framework version 2

Exact words

> "Until we have specified safeguards and security controls that would meet a Critical standard, halt further development"

Source [Preparedness Framework, Version 2](https://cdn.openai.com/pdf/18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdf) OpenAI · Apr 15, 2025

**In progress.** OpenAI's GPT-6 Astra System Card (Sep 3, 2026) rates the model Critical in cybersecurity. FLI reported on Jul 14, 2026 that OpenAI has rolled back commitments through its updated framework.

- [GPT-6 Astra System Card](https://deploymentsafety.openai.com/gpt-6-astra) OpenAI · Sep 3, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Not met[xAI](https://enterrsi.si/promises/xai/)Feb 10, 2025

### Publish a revised safety framework within three months of the February 2025 draft.

xAI, in its draft Risk Management Framework

Exact words

> "We plan to release an updated version of this policy within three months."

Source [xAI Risk Management Framework (Draft)](https://x.ai/documents/2025.02.10-RMF-Draft.pdf) xAI · Feb 10, 2025

**Not met:** reported by TechCrunch on May 13, 2025: the May 10 deadline "came and went without acknowledgement on xAI's official channels." METR lists the first full version from August 2025.

[xAI's promised safety report is MIA](https://techcrunch.com/2025/05/13/xais-promised-safety-report-is-mia/) TechCrunch · May 13, 2025

- [Frontier AI Safety Policies (index of published company frameworks)](https://metr.org/fsp) METR · Nov 10, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Withdrawn[Meta](https://enterrsi.si/promises/meta/)Feb 3, 2025

### Stop developing a model that reaches the critical risk threshold if the risk cannot be mitigated.

Meta, Frontier AI Framework

Exact words

> "If a frontier AI is assessed to have reached the critical risk threshold and cannot be mitigated, we will stop development"

Source [Frontier AI Framework](https://ai.meta.com/static-resource/meta-frontier-ai-framework) Meta · Feb 3, 2025

**Withdrawn:** reported by the Future of Life Institute on Jul 14, 2026: Meta "has also backslided on pause commitments". Its April 2026 revision changed the Critical response from Stop to Develop with Mitigations.

[AI Safety Index, Summer 2026 (full report)](https://futureoflife.org/wp-content/uploads/2026/07/AI-Safety-Index-Summer-2026-Digital-14-Jul.pdf) Future of Life Institute · Jul 14, 2026

- [Meta Superintelligence Scaling Framework, Version 2.1](https://research.meta.ai/static/meta-superintelligence-scaling-framework) Meta · Oct 2, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[Alibaba (Qwen)](https://enterrsi.si/promises/alibaba/)Dec 1, 2024

### Publicly disclose its safety and security governance measures.

Alibaba, a signatory of China's AI Safety Commitments

Exact words

> "Proactively disclose safety and security governance measures"

Source [Artificial Intelligence Safety Commitments](https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng) China AI Industry Alliance (AIIA) and CAICT · Dec 1, 2024

**No public evidence yet.** FLI reported on Jul 14, 2026 that the company has no published safety framework.

- [Artificial Intelligence Safety Commitments (China AI Industry Alliance), text and signatories](https://agora.eto.tech/instrument/1831) AGORA, Emerging Technology Observatory · Dec 1, 2024

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[Alibaba (Qwen)](https://enterrsi.si/promises/alibaba/)Dec 1, 2024

### Red-team and rigorously test models before each release or update.

Alibaba, a signatory of China's AI Safety Commitments

Exact words

> "Through dedicated simulation and red-teaming experts to rigorously test AI models prior to their release or update"

Source [Artificial Intelligence Safety Commitments](https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng) China AI Industry Alliance (AIIA) and CAICT · Dec 1, 2024

**No public evidence yet.** FLI reported on Jul 14, 2026 that it found no public mention of outside pre-release testing.

- [Artificial Intelligence Safety Commitments (China AI Industry Alliance), text and signatories](https://agora.eto.tech/instrument/1831) AGORA, Emerging Technology Observatory · Dec 1, 2024

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[DeepSeek](https://enterrsi.si/promises/deepseek/)Dec 1, 2024

### Publicly disclose its safety and security governance measures.

DeepSeek, a signatory of China's AI Safety Commitments

Exact words

> "Proactively disclose safety and security governance measures"

Source [Artificial Intelligence Safety Commitments](https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng) China AI Industry Alliance (AIIA) and CAICT · Dec 1, 2024

**No public evidence yet.** FLI reported on Jul 14, 2026 that the company has no published safety framework.

- [DeepSeek and other Chinese firms converge with Western companies on AI promises](https://carnegieendowment.org/research/2025/01/deepseek-and-other-chinese-firms-converge-with-western-companies-on-ai-promises?lang=en) Carnegie Endowment for International Peace · Jan 28, 2025

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[DeepSeek](https://enterrsi.si/promises/deepseek/)Dec 1, 2024

### Red-team and rigorously test models before each release or update.

DeepSeek, a signatory of China's AI Safety Commitments

Exact words

> "Through dedicated simulation and red-teaming experts to rigorously test AI models prior to their release or update"

Source [Artificial Intelligence Safety Commitments](https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng) China AI Industry Alliance (AIIA) and CAICT · Dec 1, 2024

**No public evidence yet.** FLI reported on Jul 14, 2026 that it found no public mention of outside pre-release testing.

- [DeepSeek and other Chinese firms converge with Western companies on AI promises](https://carnegieendowment.org/research/2025/01/deepseek-and-other-chinese-firms-converge-with-western-companies-on-ai-promises?lang=en) Carnegie Endowment for International Peace · Jan 28, 2025

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[Z.ai (Zhipu)](https://enterrsi.si/promises/zai/)Dec 1, 2024

### Publicly disclose its safety and security governance measures.

Zhipu (Beijing Zhipu Huazhang Technology), a signatory of China's AI Safety Commitments

Exact words

> "Proactively disclose safety and security governance measures"

Source [Artificial Intelligence Safety Commitments](https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng) China AI Industry Alliance (AIIA) and CAICT · Dec 1, 2024

**No public evidence yet.** FLI reported on Jul 14, 2026 that the company has no published safety framework.

- [Artificial Intelligence Safety Commitments (China AI Industry Alliance), text and signatories](https://agora.eto.tech/instrument/1831) AGORA, Emerging Technology Observatory · Dec 1, 2024

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[Z.ai (Zhipu)](https://enterrsi.si/promises/zai/)Dec 1, 2024

### Red-team and rigorously test models before each release or update.

Zhipu (Beijing Zhipu Huazhang Technology), a signatory of China's AI Safety Commitments

Exact words

> "Through dedicated simulation and red-teaming experts to rigorously test AI models prior to their release or update"

Source [Artificial Intelligence Safety Commitments](https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng) China AI Industry Alliance (AIIA) and CAICT · Dec 1, 2024

**No public evidence yet.** In its answer to FLI's survey (Jul 14, 2026), Z.ai said no outside pre-release testing was commissioned.

- [Artificial Intelligence Safety Commitments (China AI Industry Alliance), text and signatories](https://agora.eto.tech/instrument/1831) AGORA, Emerging Technology Observatory · Dec 1, 2024

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[Anthropic](https://enterrsi.si/promises/anthropic/)Aug 29, 2024

### Give the US AI Safety Institute access to major new models before and after public release.

Anthropic, in an agreement with the US AI Safety Institute

Exact words

> "access to major new models from each company prior to and following their public release"

Source [U.S. AI Safety Institute signs agreements regarding AI safety research, testing and evaluation with Anthropic and OpenAI](https://nist.gov/news-events/news/2024/08/us-ai-safety-institute-signs-agreements-regarding-ai-safety-research) NIST · Aug 29, 2024

**Kept:** reported by NIST, which published a joint US and UK pre-deployment test of the upgraded Claude 3.5 Sonnet.

[US AISI and UK AISI Joint Pre-Deployment Test: Anthropic's Claude 3.5 Sonnet (October 2024 Release)](https://www.nist.gov/document/us-aisi-uk-aisi-joint-testing-report-upgrade-claude-35-sonnet-111924) NIST · Nov 19, 2024

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[OpenAI](https://enterrsi.si/promises/openai/)Aug 29, 2024

### Give the US AI Safety Institute access to major new models before and after public release.

OpenAI, in an agreement with the US AI Safety Institute

Exact words

> "access to major new models from each company prior to and following their public release"

Source [U.S. AI Safety Institute signs agreements regarding AI safety research, testing and evaluation with Anthropic and OpenAI](https://nist.gov/news-events/news/2024/08/us-ai-safety-institute-signs-agreements-regarding-ai-safety-research) NIST · Aug 29, 2024

**Kept:** reported by NIST, which published a joint US and UK pre-deployment test of o1.

[US AISI and UK AISI Joint Pre-Deployment Test: OpenAI o1](https://www.nist.gov/document/aisi-pre-deployment-evaluation-openai-o1-december-2024) NIST · Dec 18, 2024

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[Amazon](https://enterrsi.si/promises/amazon/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

Amazon, a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Kept:** reported by METR, whose index lists the Frontier Model Safety Framework from February 2025.

[Frontier AI Safety Policies (index of published company frameworks)](https://metr.org/fsp) METR · Nov 10, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[Anthropic](https://enterrsi.si/promises/anthropic/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

Anthropic, a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Kept:** reported by METR, whose index lists the Responsible Scaling Policy from September 2023.

[Frontier AI Safety Policies (index of published company frameworks)](https://metr.org/fsp) METR · Nov 10, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[Google DeepMind](https://enterrsi.si/promises/google-deepmind/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

Google, a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Kept:** reported by METR, whose index lists the Frontier Safety Framework from May 2024.

[Frontier AI Safety Policies (index of published company frameworks)](https://metr.org/fsp) METR · Nov 10, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[Meta](https://enterrsi.si/promises/meta/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

Meta, a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Kept:** reported by METR, whose index lists the Frontier AI Framework from February 2025.

[Frontier AI Safety Policies (index of published company frameworks)](https://metr.org/fsp) METR · Nov 10, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[Microsoft](https://enterrsi.si/promises/microsoft/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

Microsoft, a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Kept:** reported by METR, whose index lists the Frontier Governance Framework from February 2025.

[Frontier AI Safety Policies (index of published company frameworks)](https://metr.org/fsp) METR · Nov 10, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Not met[Mistral](https://enterrsi.si/promises/mistral/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

Mistral AI, a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Not met:** reported by the Future of Life Institute on Jul 14, 2026: Mistral has no published safety framework.

[AI Safety Index, Summer 2026 (full report)](https://futureoflife.org/wp-content/uploads/2026/07/AI-Safety-Index-Summer-2026-Digital-14-Jul.pdf) Future of Life Institute · Jul 14, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[OpenAI](https://enterrsi.si/promises/openai/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

OpenAI, a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Kept:** reported by METR, whose index lists the Preparedness Framework from December 2023.

[Frontier AI Safety Policies (index of published company frameworks)](https://metr.org/fsp) METR · Nov 10, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[xAI](https://enterrsi.si/promises/xai/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

xAI, a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Kept:** reported by METR, whose index lists a draft Risk Management Framework from February 2025 and the first full version from August 2025.

[Frontier AI Safety Policies (index of published company frameworks)](https://metr.org/fsp) METR · Nov 10, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Not met[Z.ai (Zhipu)](https://enterrsi.si/promises/zai/)May 21, 2024

### Publish a safety framework for severe risks before the February 2025 summit in France.

Zhipu.ai (now Z.ai), a signatory of the Seoul commitments

Exact words

> "publishing a safety framework focused on severe risks by the upcoming AI Summit in France"

Source [Frontier AI Safety Commitments, AI Seoul Summit 2024](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) UK Department for Science, Innovation and Technology · May 21, 2024

**Not met:** reported by the Future of Life Institute on Jul 14, 2026: "The company has no published framework."

[AI Safety Index, Summer 2026 (full report)](https://futureoflife.org/wp-content/uploads/2026/07/AI-Safety-Index-Summer-2026-Digital-14-Jul.pdf) Future of Life Institute · Jul 14, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Amazon](https://enterrsi.si/promises/amazon/)Jul 21, 2023

### Test models inside and outside the company before release.

Amazon, one of seven companies at the White House

Exact words

> "internal and external security testing of their AI systems before their release"

Source [Fact sheet: Biden-Harris Administration secures voluntary commitments from leading artificial intelligence companies to manage the risks posed by AI](https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai) The White House (archived) · Jul 21, 2023

**In progress.** An independent CBRN uplift study for Nova 2.0 Lite is described in an Amazon paper (Jan 27, 2026); no government institute test was found.

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[Anthropic](https://enterrsi.si/promises/anthropic/)Jul 21, 2023

### Test models inside and outside the company before release.

Anthropic, one of seven companies at the White House

Exact words

> "internal and external security testing of their AI systems before their release"

Source [Fact sheet: Biden-Harris Administration secures voluntary commitments from leading artificial intelligence companies to manage the risks posed by AI](https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai) The White House (archived) · Jul 21, 2023

**Kept:** reported by NIST: the US and UK safety institutes tested the upgraded Claude 3.5 Sonnet before its October 2024 release.

[US AISI and UK AISI Joint Pre-Deployment Test: Anthropic's Claude 3.5 Sonnet (October 2024 Release)](https://www.nist.gov/document/us-aisi-uk-aisi-joint-testing-report-upgrade-claude-35-sonnet-111924) NIST · Nov 19, 2024

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Google DeepMind](https://enterrsi.si/promises/google-deepmind/)Jul 21, 2023

### Test models inside and outside the company before release.

Google, one of seven companies at the White House

Exact words

> "internal and external security testing of their AI systems before their release"

Source [Fact sheet: Biden-Harris Administration secures voluntary commitments from leading artificial intelligence companies to manage the risks posed by AI](https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai) The White House (archived) · Jul 21, 2023

**In progress.** Google's own Gemini 4 Argon safety report (Oct 13, 2026) names outside testers, including NIST CAISSI; no outside body has published its own report on it.

Checked Nov 10, 2026 · [Report a correction](#corrections)

In progress[Meta](https://enterrsi.si/promises/meta/)Jul 21, 2023

### Test models inside and outside the company before release.

Meta, one of seven companies at the White House

Exact words

> "internal and external security testing of their AI systems before their release"

Source [Fact sheet: Biden-Harris Administration secures voluntary commitments from leading artificial intelligence companies to manage the risks posed by AI](https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai) The White House (archived) · Jul 21, 2023

**In progress.** FLI reported on Jul 14, 2026 that Meta used outside testers for Muse Spark before broad internal use, with publication rights not specified.

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[Microsoft](https://enterrsi.si/promises/microsoft/)Jul 21, 2023

### Test models inside and outside the company before release.

Microsoft, one of seven companies at the White House

Exact words

> "internal and external security testing of their AI systems before their release"

Source [Fact sheet: Biden-Harris Administration secures voluntary commitments from leading artificial intelligence companies to manage the risks posed by AI](https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai) The White House (archived) · Jul 21, 2023

**No public evidence yet.** The MAI-Thinking-1 technical report (Jun 2, 2026) names no outside tester of dangerous capabilities.

Checked Nov 10, 2026 · [Report a correction](#corrections)

Kept[OpenAI](https://enterrsi.si/promises/openai/)Jul 21, 2023

### Test models inside and outside the company before release.

OpenAI, one of seven companies at the White House

Exact words

> "internal and external security testing of their AI systems before their release"

Source [Fact sheet: Biden-Harris Administration secures voluntary commitments from leading artificial intelligence companies to manage the risks posed by AI](https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai) The White House (archived) · Jul 21, 2023

**Kept:** reported by the UK AI Security Institute, which tested GPT-6 Astra before its public release.

[GPT-6 Astra performs unsanctioned supply-chain attacks in simulations](https://www.aisi.gov.uk/blog/gpt-6-astra-performs-unsanctioned-supply-chain-attacks-in-simulations) UK AI Security Institute · Sep 28, 2026

Checked Nov 10, 2026 · [Report a correction](#corrections)

Not met[OpenAI](https://enterrsi.si/promises/openai/)Jul 5, 2023

### Give 20% of its compute to the Superalignment team's work on controlling superintelligence.

OpenAI, in a post by Jan Leike and Ilya Sutskever

Exact words

> "dedicating 20% of the compute we've secured to date over the next four years"

Source [Introducing Superalignment](https://openai.com/index/introducing-superalignment/) OpenAI · Jul 5, 2023

**Not met:** reported by Fortune on May 21, 2024: "OpenAI never fulfilled its commitment to provide the team with 20% of its computing power."

[OpenAI promised 20% of its computing power to combat the most dangerous kind of AI, but never delivered, sources say](https://fortune.com/2024/05/21/openai-superalignment-20-compute-commitment-never-fulfilled-sutskever-leike-altman-brockman-murati/) Fortune · May 21, 2024

Checked Nov 10, 2026 · [Report a correction](#corrections)

Withdrawn[Google DeepMind](https://enterrsi.si/promises/google-deepmind/)Jun 7, 2018

### Not pursue weapons or other technologies whose main purpose is to injure people.

Google, in "AI at Google: our principles" by CEO Sundar Pichai

Exact words

> "Weapons or other technologies whose principal purpose or implementation is to cause or directly facilitate injury to people."

Source [AI at Google: our principles](https://blog.google/technology/ai/ai-principles/) Google · Jun 7, 2018

**Withdrawn:** reported by TechCrunch on Feb 4, 2025: Google removed the "applications we will not pursue" section, weapons included, from its published principles.

[Google removes pledge to not use AI for weapons from website](https://techcrunch.com/2025/02/04/google-removes-pledge-to-not-use-ai-for-weapons-from-website/) TechCrunch · Feb 4, 2025

Checked Nov 10, 2026 · [Report a correction](#corrections)

No public evidence yet[OpenAI](https://enterrsi.si/promises/openai/)Apr 9, 2018

### Stop competing with, and start assisting, a safety-conscious project that comes close to building AGI first.

OpenAI, in its Charter

Exact words

> "we commit to stop competing with and start assisting this project"

Source [OpenAI Charter](https://openai.com/charter/) OpenAI · Apr 9, 2018

**No public evidence yet.** The clause is still on the Charter page; no public report says it has been invoked.

Checked Nov 10, 2026 · [Report a correction](#corrections)

The method

## Five statuses, all of them sourced or cautious

A promise is listed only with a primary source. The status is the strongest one a reputable public source supports, and nothing stronger.

Status definitions

### What each status means

- Kept
  A reputable public source reports the promise fulfilled. The source is linked and named, with its date.
- Not met
  A reputable public source reports that the promise was not fulfilled, or that a deadline passed without it.
- Withdrawn
  The promise was removed or replaced by the one who made it, as a reputable public source reports.
- In progress
  The promise is ongoing or not yet due, and public records show work toward it.
- No public evidence yet
  No public, reputable source shows the promise kept or broken. This is the cautious default.

Corrections

### Spotted an error?

Every entry links to its sources and the day it was checked. If a promise is missing, misquoted or has a newer status, send the public source on [Telegram](https://t.me/enterrsitg) or [X](https://x.com/enterrsi). Corrections are made in the open, with the new check date.

[Send a correction](https://t.me/enterrsitg)

Cite the Promise Tracker

The Promise Tracker: public safety promises of SI labs. enterrsi.si, checked November 10, 2026. 47 promises, 11 labs. https://enterrsi.si/promises/

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

[Download JSON](https://enterrsi.si/promises/promises.json)[Download CSV](https://enterrsi.si/promises/promises.csv)

Embed

### Put this tally on your site

A small card with the overall promise tally and a link back. It updates with each release of this page.

```
<div data-rsi-promises></div><script src="https://enterrsi.si/promises/embed.js" async></script>
```

Pass it on

### Share the tracker

Questions

## Straight answers

How the tracker works, in four answers.

### What counts as a promise?

A public, forward-looking commitment about safety made by a lab or its leader: in a signed public commitment such as the 2023 White House voluntary commitments or the 2024 Seoul Frontier AI Safety Commitments, in testimony, in a published framework or on the lab's own pages. Each one is listed only with that primary source.

### Who decides whether a promise was kept?

Not this site. Kept, Not met and Withdrawn appear only when a reputable public source reports it, and the entry names that source and its date. Otherwise the status is In progress or No public evidence yet, and when in doubt the cautious status is used.

### How does this relate to the Containment Board?

The [Containment Board](https://enterrsi.si/containment-board/) grades what each lab has in place today. The Promise Tracker follows what each lab said it would do. Each lab page links to the other.

### How do I correct an entry?

Send the public source on [Telegram](https://t.me/enterrsitg) or [X](https://x.com/enterrsi). Corrections are made in the open, and the entry gets a new check date.

Enter RSI

## Promises on one side, grades on the other

The Containment Board grades what each lab has in place today, from safety frameworks to whistleblower policies.

[Open the Containment Board](https://enterrsi.si/containment-board/)
[The Control Map](https://enterrsi.si/control-map/)
[The Comment Desk](https://enterrsi.si/comment-desk/)