## Research

### Publication

- **Apr 29, 2026**  
  [Where the goblins came from](/content/index/where-the-goblins-came-from/index.html)  
  How goblin outputs spread in AI models: timeline, root cause, and fixes behind personality-driven quirks in GPT-5 behavior.

- **Apr 23, 2026**  
  [GPT-5.5 System Card](/content/index/gpt-5-5-system-card/index.html)

- **Mar 25, 2026**  
  [Inside our approach to the Model Spec](/content/index/our-approach-to-the-model-spec/index.html)  
  Learn how OpenAI’s Model Spec serves as a public framework for model behavior, balancing safety, user freedom, and accountability as AI systems advance.

- **Mar 19, 2026**  
  [How we monitor internal coding agents for misalignment](/content/index/how-we-monitor-internal-coding-agents-misalignment/index.html)  
  How OpenAI uses chain-of-thought monitoring to study misalignment in internal coding agents—analyzing real-world deployments to detect risks and strengthen AI safety safeguards.

- **Mar 10, 2026**  
  [Improving instruction hierarchy in frontier LLMs](/content/index/instruction-hierarchy-challenge/index.html)  
  IH-Challenge trains models to prioritize trusted instructions, improving instruction hierarchy, safety steerability, and resistance to prompt injection attacks.

- **Mar 5, 2026**  
  [Reasoning models struggle to control their chains of thought, and that’s good](/content/index/reasoning-models-chain-of-thought-controllability/index.html)  
  OpenAI introduces CoT-Control and finds reasoning models struggle to control their chains of thought, reinforcing monitorability as an AI safety safeguard.

- **Mar 5, 2026**  
  [GPT-5.4 Thinking System Card](/content/index/gpt-5-4-thinking-system-card/index.html)

- **Mar 4, 2026**  
  [Extending single-minus amplitudes to gravitons](/content/index/extending-single-minus-amplitudes-to-gravitons/index.html)  
  A new preprint extends single-minus amplitudes to gravitons, with GPT-5.2 Pro helping derive and verify nonzero graviton tree amplitudes in quantum gravity.

- **Mar 3, 2026**  
  [GPT-5.3 Instant System Card](/content/index/gpt-5-3-instant-system-card/index.html)