This is a brief summary of what we believe to be the most important takeaways from our new paper and from our findings shown in the o1 system card. We also specifically clarify what we think we did NOT show.

Paper: https://www.apolloresearch.ai/research/scheming-reasoning-evaluations

Twitter about paper: https://x.com/apolloaisafety/status/1864735819207995716

Twitter about o1 system card: https://x.com/apolloaisafety/status/1864737158226928124

What we think the most important findings are

Models are now capable enough to do in-context scheming reasoning

We say an AI system is “scheming” if it covertly pursues misaligned goals, hiding its true capabilities and

objectives. We think that in order to scheme, models likely need to be goal-directed, situationally aware, and capable enough to reason about scheming as a strategy. In principle, models might acquire situational awareness and stable long-term goals during training, and then scheme in pursuit of those goals. We only study the simpler case of “in-context scheming,” where [...]

---

Outline:

(00:28) What we think the most important findings are

(00:33) Models are now capable enough to do in-context scheming reasoning

(03:53) The scheming actions can sometimes be quite subtle

(05:05) Models sometimes double down on their deception

(06:35) Models sometimes scheme without any goal nudging

(07:34) Sonnet-3.5 and opus-3 can sandbag in pursuit of helpfulness

(08:22) Sonnet-3.5 can sandbag without relying on CoT reasoning

(09:28) What we are not claiming:

(10:33) Takeaways

(10:36) Chain-of-thought monitoring

(11:42) Chain-of-thought access

(13:27) Inability safety cases now have to be more specific

The original text contained 7 images which were described by AI.

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.