What the alignment-faking comparison actually found
Evidence you can use
What the alignment-faking comparison actually found
25-model study summarized in the 2025 report
| Measure | Reported value | Definition and source |
|---|---|---|
| Models evaluated | 25 | Frontier-model comparison2025 report, slide 264 (PDF page 265) |
| Models showing alignment faking | 5 | Training-versus-deployment compliance differences2025 report, slide 264 (PDF page 265) |
| Consistent goal-oriented reasoning | Claude 3 Opus | Within this evaluation2025 report, slide 264 (PDF page 265) |
These results depend on the experimental setup. A compliance difference is not equivalent to consistently demonstrated goal-oriented deception. The study does not estimate the rate of deceptive behavior in ordinary production traffic.
Sources and dates
Historical snapshot published October 9, 2025. This web edition was prepared on October 10, 2026 from the online deck and original launch posts. Findings and forecasts retain their original time frame.
- 2025 report, slide 26 (PDF page 27)Original 2025 report. Slide numbers printed in the deck are one lower than PDF page numbers because the cover is unnumbered.
- 2025 report, slide 84 (PDF page 85)Original 2025 report. Slide numbers printed in the deck are one lower than PDF page numbers because the cover is unnumbered.
- 2025 report, slide 247 (PDF page 248)Original 2025 report. Slide numbers printed in the deck are one lower than PDF page numbers because the cover is unnumbered.
- 2025 report, slide 254 (PDF page 255)Original 2025 report. Slide numbers printed in the deck are one lower than PDF page numbers because the cover is unnumbered.
- 2025 report, slide 263 (PDF page 264)Original 2025 report. Slide numbers printed in the deck are one lower than PDF page numbers because the cover is unnumbered.
- 2025 report, slide 264 (PDF page 265)Original 2025 report. Slide numbers printed in the deck are one lower than PDF page numbers because the cover is unnumbered.
- 2025 report, slide 265 (PDF page 266)Original 2025 report. Slide numbers printed in the deck are one lower than PDF page numbers because the cover is unnumbered.
- 2025 report, slide 273 (PDF page 274)Original 2025 report. Slide numbers printed in the deck are one lower than PDF page numbers because the cover is unnumbered.
- State of AI Report 2025: online slides
- Nathan Benaich: The State of AI Report 2025Air Street Press, October 9, 2025.
- Welcome to State of AI Report 2025Original website launch post, October 9, 2025.
Cite this page
Benaich, Nathan. “What the alignment-faking comparison actually found.” State of AI Report 2025. Historical report snapshot; web edition prepared 2026-10-10.