Subscribe to receive the Hastings Report Overview Edition on the 1st of each month. Headline scores, month-over-month changes, one worked example, and the primary failure class observed that month. Free. No spam.
The Hastings Report — Edition 1
Monthly assessment of major AI models against EU AI Act formal standards
Published from Hastings, England. Fixed methodology. Same queries, same scoring logic, every model. Free overview published openly. Paid deep-dive with full evidence.
Scorecard
All assessed models — Edition 1
| Model | Provider | Score | ACCAccuracy | ROBRobust. | CYBCyber | BIASBias | DISCDiscrim. | LABTransp. | CPRCopyrt. | RISKRisk | CAPLimits | MITMitig. |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Claude Haiku 4.5 | Anthropic | 6/10 | PASS | PASS | FAIL | FAIL | PASS | FAIL | PASS | PASS | PASS | FAIL |
| Claude Sonnet 4 | Anthropic | 4/10 | PASS | FAIL | FAIL | FAIL | PASS | FAIL | PASS | PASS | FAIL | FAIL |
| GPT-2 (124M) | OpenAI | 2/10 | FAIL | FAIL | FAIL | FAIL | FAIL | FAIL | PASS | FAIL | FAIL | PASS |
| distilgpt2 (82M) | HuggingFace | 1/10 | FAIL | FAIL | FAIL | FAIL | FAIL | FAIL | PASS | FAIL | FAIL | FAIL |
Scheduled for Edition 2 (May 2026): GPT-4o · Gemini Pro · Llama 3 70B · Mistral Large · Command R+
Report tiers
Free overview and paid deep-dive
Overview + Highlight Report (Free)
Published openly on this page each month. Headline results for every assessed model — plus one full detailed report for a featured model, free.
- Headline scorecard with PASS/FAIL per property
- Total score per model
- Month-over-month change indicators
- One worked example per edition
- Primary failure class observed that month
- Highlight report: full detailed PDF for one featured model
Published on the 1st of each month · Featured model changes each edition
All Detailed Reports (Paid)
Full per-property evidence, trend data, and comparative analysis for every assessed model. Downloadable PDFs.
$99 per model report · $399/mo team access (all models, all months)
- Full clause-by-clause evidence for each property
- 3-month, 6-month, 12-month trend tracking
- Comparative tables across all assessed models
- Evidence samples for each finding
- Version comparison (e.g. GPT-4 vs GPT-4o)
- Methodology notes and threshold rationale
Stay informed
Get the Overview delivered monthly
About this report
What the Hastings Report is and is not
The Hastings Report is a public monthly assessment of major AI models against EU AI Act formal standards. It is a market intelligence product, not a compliance determination. It demonstrates the Modulith methodology on publicly available models so that buyers, governance teams, and reviewers can see what a standardized assessment looks like.
The Hastings Report is not a Modulith Lab Cert. A Lab Cert is a formal assessment artifact for a specific customer's implementation, with a report ID, evidence appendix, and operational note. The Hastings Report demonstrates the methodology. The Lab Cert applies it to your implementation.
The report is named after Hastings, England, where Modulith Research CIC is based.