NUWA DEVELOPER DISCLOSURE · Anthropic
Anthropic August 2026 Risk Report
Developer documentation covering Claude model family, internal models discussed in redacted report.
Published 2026-08-14
Sources ↓Developer-authored disclosure. Selected source sections were read; numerical results and comparability have not been audited. NUWA has not independently replicated these experiments.
Models covered
SHADE-Arena
Covert side-task completion and monitor detection under defined tool access.
A protocol mention does not establish cross-report comparability or independent replication.
This section is catalogued; no precise numerical result has been extracted here. Open the original report for its figures and full findings.
Report coverage
| Section | Source location | Treatment |
|---|---|---|
| 2 Autonomy threat model 1: Misalignment in high-stakes settings | PDF printed pp.17–88 | Linked protocol Structured claims, monitoring, mitigation assumptions and limitations support a low-risk developer judgment; redactions constrain independent review. |
| 3 Autonomy threat model 2: Risks from automated R&D | PDF printed pp.93–113 | Contextual evidence Separates researcher substitution, observed acceleration and forward-looking risks; capability proxies do not establish a measured economy-wide acceleration. |
| 4 Chemical and biological weapons production | PDF printed pp.115–155 | Contextual evidence Threat models, evidence weighting and mitigations reviewed at a high level only; no hazardous task details or quantitative results retained. |
| 5.2 Safety process failures | PDF printed pp.163–168 | Contextual evidence Developer discusses safety-process failures separately from capability evaluations; these are investigation/governance evidence. |
References
- [1] Anthropic August 2026 Risk Report. 2026-08-14