Alignment faking in large language models
Controlled study varies apparent training conditions and examines compliance and reasoning traces.
Verify byline in the original ↗PUBLICATION BYLINE
Indexed publications with this author name. Affiliation and author identity require publication-specific evidence.
Controlled study varies apparent training conditions and examines compliance and reasoning traces.
Verify byline in the original ↗Evaluates protocols that use trusted oversight to limit harm from a more capable model that intentionally subverts the task.
Verify byline in the original ↗Identifies chain-of-thought monitoring as a safety research opportunity and analyzes conditions under which it may become less reliable.
Verify byline in the original ↗The authors judge risk very low but nonzero and describe disagreements about evidence and argumentation.
Verify byline in the original ↗