Science – Apollo Research
https://www.apolloresearch.ai/science
Apollo Research — named in the OpenAI system card for sandbagging and scheming evaluations
7
Lines added
7
Lines removed
Was read
2026-09-23
d28e37eddcbae960…
Changed by
2026-10-15
25de38de0924e1ab…
7 lines added, 7 lines removed
Left column is the version read 2026-09-23, right is 2026-10-15. Unchanged stretches are elided. Both bodies are kept in full and are addressed by the hashes above.
Showing the first 400 lines of the comparison; 63 further lines are held back. This page reports a change, it is not a copy of anyone’s document — the source is linked above and is the authority for what it says.
⋯ 54 unchanged lines
5555 19 January 2026
5656
5757 Science of Scheming
5858
59+GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior
60+Evaluations
61+
62+15 October 2026
63+
64+Evaluations
65+
5966 Measuring Reward-Seeking via Contrastive Belief Updates
6067 Science of Scheming
6168
6269 21 July 2026
⋯ 57 unchanged lines
120127
121128 24 February 2025
122129
123130 Evaluations
124−
125−Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition
126−Interpretability
127−
128−11 February 2025
129−
130−Interpretability
131131
132132 Load more
133133
134134 Evaluations