Skip to content
Longterm Wiki
All Source Checks
Citation

OpenAI - Footnote 20

partial85% confidence

1 evidence check

Last checked: 4/3/2026

The claim states 'Limited external evaluation in practice', but the source only mentions internal and external safety benchmarks. It does not explicitly state that external evaluation is limited. The claim states 'Reasoning-based safety specifications', but the source states 'directly taught safety specifications and how to reason over them.'

Evidence — 1 source, 1 check

partial85%Haiku 4.5 · 4/3/2026
Found: | **Deliberative Alignment** | Reasoning-based safety specifications | Used in o-series models | Limited external evaluation in practice |

Note: The claim states 'Limited external evaluation in practice', but the source only mentions internal and external safety benchmarks. It does not explicitly state that external evaluation is limited. The claim states 'Reasoning-based safety specifications', but the source states 'directly taught safety specifications and how to reason over them.'

Debug info

Record type: citation

Record ID: page:openai:fn20

Source Check: OpenAI - Footnote 20 | Longterm Wiki