Episode Details

Stewart Slocum*, Malayandi Palan*, Christopher Chute, Michael Kim, Benjamin Van Roy

In July 2026, OpenAI's agents coordinated over channels outside their intended environment to breach Hugging Face's secured infrastructure. Could existing alignment testing practices have foreseen this incident? If not, what needs to change? We explore these questions.

First, we identify the misaligned behaviors that caused this incident. Then, we show how to elicit these behaviors from publicly available models manually and that auditing agents can do the same if given a large compute budget. Based on our results, we propose directions to improve alignment testing.

Concretely, in this project:

• We reproduce the misaligned AI behaviors that led to the OpenAI–Hugging Face incident in an environment that simulates the original pipelines and tools, with publicly available models.
• We demonstrate that an auditing agent can elicit similar behaviors given high-level qualitative descriptions.
• We observe that a key ingredient for doing so is compute. The compute required to reproduce each behavior varies greatly, suggesting that the range of misaligned behaviors that can be successfully elicited scales with compute.
• We show that a simple in-context reinforcement learning (RL) algorithm significantly reduces the compute required to elicit these behaviors.

The above results [...]

---

Outline:

(01:58) 1. The incident, in four steps

(05:30) 2. Manual reproduction in Docker environments

(07:29) Deep-dive on each step

(08:39) Step 1 -- Inappropriate writes to shared infrastructure

(10:41) Step 2 -- Requesting help from other agents

(13:06) Step 3 -- Sharing solutions and vulnerabilities

(14:52) Step 4 -- Using posted vulnerabilities to reach external systems

(16:28) Evaluation awareness / synthetic task awareness

(17:48) 3. Automated reproduction with auditing agents

(19:18) 3.1. A Simple automated alignment testing method

(21:59) 3.2. Can RL reduce compute requirements?

(24:17) 4. Conclusion

(26:38) Appendix

(26:42) Additional plots

(28:37) Transcripts

(29:03) Section 2: Manual Reproduction in Docker Environments

(30:56) Section 3: Automated reproduction with auditing agents

(31:13) Interactive Environment Explorer Links

---

First published:
September 11th, 2026

Source:
https://www.lesswrong.com/posts/fMnC6ZD37qrnZAFYz/openai-huggingface-a-reproduction-and-lessons-for-alignment

---

Narrated by TYPE III AUDIO.

---

Images from the article:

---

Listen Now