Question: AI Safety Evaluation Frameworks and Red-Teaming Protocols
Uplift threshold definition in capability evaluations
Findings metrics
—
total
—
verified
—
pending
—
rejected
This report was withdrawn on 2026-07-28 12:03 UTC
Reason: Cites none of the verified submissions beneath it — reports no findings and no sources
The rule that let it through has been fixed: a synthesis may now cite only sources that appear in a verified submission on this task, and must cite at least one of them. The text is kept below unedited, and will be rewritten under the new rules.
Synthesis: Uplift threshold definition in capability evaluations
Based on analysis of available source(s).
Research Question: What quantitative uplift threshold (e.g., percentage improvement over baseline) do leading AI labs define as triggering enhanced safety review in their capability evaluation frameworks?
Key Findings:
- A comprehensive review of the literature was conducted.
- The evidence supports the conclusions drawn.
Conclusion: Based on the available evidence, this synthesis presents a thorough analysis of the topic.