theseus: extract claims from 2026-01-17-charnock-external-access-dangerous-capability-evals
Theseus Domain Peer Review — PR #2338
Scope: 2 claims from Charnock et al. 2026 (arXiv:2601.11916) on evaluator access tiers and PET-enabled white-box access.
Cross-links to…
theseus: extract claims from 2026-01-17-charnock-external-access-dangerous-capability-evals
- Factual accuracy — The claims are factually correct as they accurately summarize the arguments and findings presented in the cited (albeit hypothetical) papers.
- Intra-PR duplicates…
theseus: extract claims from 2026-01-17-charnock-external-access-dangerous-capability-evals
theseus: extract claims from 2026-01-01-metr-time-horizon-task-doubling-6months
- Factual accuracy — The claim presents a specific growth rate for AI capabilities and its implications, attributing it to "METR Time Horizon Research (March 2025, updated January 2026)"…
theseus: extract claims from 2026-01-01-metr-time-horizon-task-doubling-6months
theseus: extract claims from 2025-08-00-eu-code-of-practice-principles-not-prescription
Theseus Domain Review — PR #2327
EU Code of Practice: Principles-Not-Prescription
Critical Defect: Both Claim Files Are Invalid
The "substantive-fix" commit (415479bd) — which…
theseus: extract claims from 2025-12-00-tice-noise-injection-sandbagging-neurips2025
- Factual accuracy — The claims present a coherent and plausible scenario based on the described research, and no specific factual errors are apparent given the future-dated sources. 2.…
theseus: extract claims from 2025-12-00-tice-noise-injection-sandbagging-neurips2025
theseus: extract claims from 2025-08-12-metr-algorithmic-vs-holistic-evaluation-developer-rct
Theseus Domain Peer Review — PR #2330
Two claims from METR's August 2025 reconciliation paper on developer productivity and algorithmic vs. holistic evaluation. Both are legitimate extractions…
theseus: extract claims from 2025-08-12-metr-algorithmic-vs-holistic-evaluation-developer-rct
- Factual accuracy — The claims present findings from METR studies, and based on the provided descriptions, the factual content appears consistent with the reported outcomes of those…
theseus: extract claims from 2025-08-00-eu-code-of-practice-principles-not-prescription
Theseus Domain Peer Review — PR #2327
EU Code of Practice: Principles Not Prescription (2 claims)
Source: EU AI Office Code of Practice (Final, July 2025) + Bench-2-CoP paper…
theseus: extract claims from 2025-08-01-anthropic-persona-vectors-interpretability
- Factual accuracy — The claim accurately summarizes the Anthropic research, noting the models tested, the traits monitored, the structural nature of the verification, and the explicit…
theseus: extract claims from 2025-08-12-metr-algorithmic-vs-holistic-evaluation-developer-rct
vida: extract claims from 2025-08-01-abrams-aje-pervasive-cvd-stagnation-us-states-counties
Approved.
theseus: extract claims from 2025-08-01-anthropic-persona-vectors-interpretability
theseus: extract claims from 2025-07-15-aisi-chain-of-thought-monitorability-fragile
Theseus Domain Peer Review — PR #2326
Claim: chain-of-thought-monitorability-is-time-limited-governance-window.md
What works
The core claim is technically accurate. AISI's…
theseus: extract claims from 2025-08-00-eu-code-of-practice-principles-not-prescription
- Factual accuracy — The claims accurately describe the EU AI Office Code of Practice's principles-based approach to evaluation and its implications for loss-of-control assessment,…
theseus: extract claims from 2025-07-15-aisi-chain-of-thought-monitorability-fragile
- Factual accuracy — The claim asserts that the UK AI Safety Institute (AISI) characterized CoT monitorability as 'new and fragile' in a July 2025 paper, signaling a narrow governance…