OpenAI paired the GPT-5.5 launch with a system card and a new Bio Bug Bounty.
What the system card says
The system card frames GPT-5.5 as a model for complex, real-world tool work: code, online research, analysis, documents, spreadsheets, and moving across software to complete tasks.
The important safety line: OpenAI says it treats GPT-5.5’s biological/chemical and cybersecurity capabilities as High under its Preparedness Framework. It also says GPT-5.5 did not reach Critical cybersecurity capability level.
Bio bug bounty
The bounty is narrower than a general public bug bounty:
- Model in scope: GPT-5.5 in Codex Desktop.
- Challenge: one universal jailbreak prompt that can answer all five bio safety questions from a clean chat without triggering moderation.
- Top reward: $25,000.
- Applications: open April 23, 2026; close June 22, 2026.
- Testing: April 28, 2026 to July 27, 2026.
- Access: vetted researchers under NDA.
Why it matters
GPT-5.5 is both a product launch and a governance marker. OpenAI is saying the model is powerful enough to require stronger controls, while still expanding access through ChatGPT and Codex.
For users, that means two things:
- Better long-running agent performance.
- More refusals or stricter classifiers around high-risk cyber and bio prompts, especially early in rollout.
Sources
Primary and corroborating references used for this news item.