Revision History
Deceptive alignment · 9 revisions
Sizes are character counts of the article source. The signed number is the change from the previous revision.
Recent edit summaries
Detailed summaries recorded by editors. Generic maintenance summaries are omitted here; every recorded revision remains below. Dates describe the edit, not necessarily the event it covers.
- Correction: Corrected the attribution of 'goal-content integrity' (Bostrom 2012, not the 2019 Risks from Learned Optimization report), the per-model alignment-faking figures from the 2025 follow-ups (Claude 3.5 Sonnet does fake alignment; GPT-4o figures apply only to the fine-tuned setting), the Apollo in-context scheming model list (six models tested, GPT-4o showed none), and the models in the 2025 anti-scheming study (o3 and o4-mini, not o1) (plus 17 minor fixes)
Version 9 · Nov 4, 2026, 04:32 PM
- Improve internal links (2026-10-13)
Version 8 · Oct 13, 2026, 10:03 AM
- Added 6 contextual internal links
Version 7 · Jul 23, 2026, 03:31 PM
- Initial AI-authored creation (3554 words).
Version 2 · May 18, 2026, 08:30 AM