Certified AI cloud, but hardly any rules against misuse of computing capacity
Overall rating “limited”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Only a general oversight structure is documented; there is no AI policy, no approval procedure, no review of AI customers and no responsible AI report.
What is documented
The case against
No AI rules or responsibilities
No accountability
- No responsible AI or sustainability report was found; ISO/IEC 42001 is only being pursued10
Sources for Responsibility (10)
- CoreWeave, Inc. (Investor Relations): Definitive Proxy Statement 2026 Annual Meeting (PDF) · 22 April 2026 · primary source
- CoreWeave, Inc. (SEC EDGAR): Form DEF 14A (Proxy Statement 2026) · 22 April 2026 · primary source
- Boardroom Alpha: CRWV DEF 14A 2026-04-22 (summary) · 22 April 2026 · secondary source
- ROI-NJ: CoreWeave announces new chief information security officer Jim Higgins · 27 February 2025 · secondary source
- CoreWeave, Inc. (SEC EDGAR): Form 10-K for the fiscal year ended December 31, 2025 · 2 March 2026 · primary source
- CoreWeave, Inc.: Acceptable Use Policy · 17 September 2021 · secondary source
- CoreWeave, Inc.: Privacy Policy · 24 February 2026 · secondary source
- CoreWeave, Inc.: Code of Business Conduct and Ethics (approved 2026-09-10) · 10 September 2026 · secondary source
- CoreWeave, Inc. (SEC EDGAR): Form 8-K, Item 5.07 voting results of the annual general meeting of 2026-06-08 · 8 June 2026 · primary source
- CoreWeave, Inc. (Blog, James Higgins): How CoreWeave Builds Security Into the Architecture That Powers Modern AI · 18 November 2025 · secondary source
2 · Safety
The rating rests essentially on the certifications cited in the 10-K, whose reports are available only under confidentiality, while protection against misuse is hardly regulated.
What is documented
- The 10-K cites SOC 2 and ISO/IEC 27001 certifications; the Trust Center additionally lists ISO 27017 and 270184 1
- According to an account of Item 1C, there are a risk register, monitoring, penetration testing and assessments by independent firms5
- SOC 2 Type II for Bare Metal and CKS, DPU-based isolation and immutable logs are described7
- The Zero Trust AI Framework provides customer-managed keys and MACsec; confidential computing has only been announced8
The case against
Weak protection against misuse
- The 2021 acceptable use policy prohibits no AI-specific uses such as deepfakes, weapons or surveillance and names no reporting channel9
- No abuse monitoring for AI workloads and no enforcement figures are described
Limited verifiability
Sources for Safety (9)
- CoreWeave, Inc.: CoreWeave Trust Center · 3 November 2026 · secondary source
- Weights & Biases (CoreWeave): W&B Weave Guardrails · 3 November 2026 · secondary source
- IsDown (status aggregator): CoreWeave Serverless and W&B Inference request failures incident · 31 July 2026 · secondary source
- CoreWeave, Inc. (SEC EDGAR): Form 10-K for the fiscal year ended December 31, 2025 · 2 March 2026 · primary source
- Board Cybersecurity: CoreWeave, Inc. 10-K Item 1C. Cybersecurity – 2026-03-02 · 2 March 2026 · secondary source
- Weights & Biases (CoreWeave): Security (W&B) · 3 November 2026 · secondary source
- CoreWeave, Inc. (Blog, James Higgins): How CoreWeave Builds Security Into the Architecture That Powers Modern AI · 18 November 2025 · secondary source
- CoreWeave, Inc.: CoreWeave Launches the CoreWeave Zero Trust AI Framework · 21 October 2026 · secondary source
- CoreWeave, Inc.: Acceptable Use Policy · 17 September 2021 · secondary source
3 · Transparency
CoreWeave commits not to use customer data for its own models, yet its privacy policy allows inputs and outputs to be used for product development, and information on the services' limits is missing.
What is documented
- The 10-K contains AI risk factors; the privacy policy mentions the processing of personal data using AI and ML1 2
- CoreWeave commits not to use customer prompts, training data, weights or outputs for its own models5
- The W&B Weave guardrail scorers against toxicity, bias, personal data, hallucinations and prompt injection are described3
The case against
Unclear data use
Missing information on limitations
- There is no information on the accuracy and limits of the guardrail scorers and inference services3
- No transparency report on government requests and abuse cases was found
Sources for Transparency (5)
- CoreWeave, Inc. (SEC EDGAR): Form 10-K for the fiscal year ended December 31, 2025 · 2 March 2026 · primary source
- CoreWeave, Inc.: Privacy Policy · 24 February 2026 · secondary source
- Weights & Biases (CoreWeave): W&B Weave Guardrails · 3 November 2026 · secondary source
- IsDown (status aggregator): CoreWeave Serverless and W&B Inference request failures incident · 31 July 2026 · secondary source
- CoreWeave, Inc.: CoreWeave Launches the CoreWeave Zero Trust AI Framework · 21 October 2026 · secondary source
4 · Why the harm rule does not apply
The harm rule does not apply: there is no judgment, no fine and no pattern of serious incidents involving its own AI systems; the only recorded incident is a brief disruption of the inference services.
5 · Controversies (1)
6 · What remains open
- Responsibility for AI risks at board or management level
- AI policy, risk review of AI customers and abuse monitoring
- Original text of Items 1C and 3 of the 10-K
- Audit reports and scope of the SOC 2 and ISO certifications
- Testing, limits and error rates of the inference and guardrail services
7 · How it was assessed
Cut-off date: 3 November 2026 · 40 searches, 33 fetches.
Primary documents read (6)
- Form 10-K FY2025 (cover page, Item 1, Item 1A; two retrievals)
- Definitive Proxy Statement 2026 (company PDF, two retrievals; SEC version DEF 14A, one retrieval)
- Form 10-Q Q2 2026 (notes up to and including leases; two retrievals)
- Form 8-K voting results AGM 2026-06-08; Form 8-K earnings release Q1 2026; Form 8-K Exhibit 99.1 convertible notes 2026-09-17
- Code of Business Conduct and Ethics (2026-09-10), Acceptable Use Policy (2021-09-17), Privacy Policy (2026-02-24), Trust Center (current version and archived version)
- W&B Security and Guardrails; CISO Security Blog (2025-11-18)
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0