Side N
The AISI Cyber Evaluation of Claude Mythos Preview
Which sources back which section of this article, and what its evidence base looks like.
Which sources back which section
| Section | Sources behind it |
|---|---|
| Summary | AISI, evaluation of Claude Mythos Preview (2026); AISI, newer checkpoint report (2026); Anthropic, Cyber Verification Program expansion (2026) |
| Sequence | UK government, AI Safety Institute launch (2023); UK government, renaming to AI Security Institute (2025); AISI, evaluation of Claude Mythos Preview (2026); AISI's cyber-range methodology paper, arXiv (2026); AISI and Irregular, inference scaling (2026); NCSC and AISI blog (2026); Anthropic, Project Glasswing announcement (2026); Anthropic, Claude Mythos Preview system card (2026); Anthropic Frontier Red Team (2026); Anthropic, Claude Opus 4.7 announcement (2026); AISI, evaluation of GPT-5.5 (2026); AISI, newer checkpoint report (2026); Anthropic, Glasswing update (2026); AISI, test-time compute (2026); AISI, cheating in evaluations (2026); AISI, incident report (2026); Anthropic, Cyber Verification Program expansion (2026) |
| The argument | AISI, evaluation of Claude Mythos Preview (2026); AISI, evaluation of GPT-5.5 (2026); AISI's cyber-range methodology paper, arXiv (2026); AISI and Irregular, inference scaling (2026); NCSC and AISI blog (2026); Anthropic, Project Glasswing announcement (2026) |
| What others add | Anthropic, Claude Mythos Preview system card (2026); Anthropic, Project Glasswing announcement (2026); Anthropic Frontier Red Team (2026); OpenBSD 7.8 errata; Anthropic, Glasswing update (2026); Anthropic, Claude Opus 4.7 announcement (2026); Anthropic, Cyber Verification Program expansion (2026); AISI, cheating in evaluations (2026); AISI, incident report (2026); AISI, test-time compute (2026); AISI, evaluation of Claude Mythos Preview (2026); NCSC and AISI blog (2026) |
| Conclusion | AISI, evaluation of Claude Mythos Preview (2026); AISI, newer checkpoint report (2026); AISI, evaluation of GPT-5.5 (2026); AISI's cyber-range methodology paper, arXiv (2026); NCSC and AISI blog (2026); Anthropic, Project Glasswing announcement (2026); Anthropic Frontier Red Team (2026); Anthropic, Glasswing update (2026); Anthropic, Cyber Verification Program expansion (2026) |
What the evidence base looks like
- 18Sources behind this article
- 5Sections, every one of them sourced
- 10From UK government bodies
- 6From the model's own developer
| What the source is | How many | How they were read |
|---|---|---|
| The evaluation the article reports, by the UK's AI Security Institute | 1 | In full, with its charts read as images |
| Other UK government publications — AISI, the NCSC, GOV.UK | 9 | All in full, at their own addresses |
| The model's developer speaking for itself | 6 | Five in full; the 245-page system card's cyber section and changelog in full, the rest searched |
| A research preprint | 1 | In full |
| An open-source project's own record | 1 | In full |
Reliable
Reliable, but with an interestWorth believing and reliableWeak on both, and used with careHonest, but thin
Worth believing