GPT-5.5 Bio Bug Bounty

2026-04-23 · Source: OpenAI News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, quick

Summary

OpenAI has launched a Bio Bug Bounty program for its GPT-5.5 model, specifically within the Codex Desktop environment, to identify universal jailbreaks for biorisk safeguards. The program invites researchers experienced in AI red teaming, security, or biosecurity to find a single prompt that can bypass the model's moderation and successfully answer all five bio safety challenge questions. The first successful universal jailbreak will receive a $25,000 reward, with smaller awards possible for partial successes. Applications opened on April 23, 2026, and close on June 22, 2026, with testing running from April 28, 2026, to July 27, 2026. All findings and communications are subject to an NDA.

Key takeaway

For AI scientists and security researchers focused on frontier model safety, participating in OpenAI's GPT-5.5 Bio Bug Bounty offers a direct opportunity to contribute to critical biorisk mitigation. Your expertise in identifying universal jailbreaks is crucial for strengthening safeguards, and successful contributions are financially rewarded. Consider applying by the June 22, 2026 deadline to help secure advanced AI capabilities.

Key insights

OpenAI seeks universal jailbreaks for GPT-5.5's biorisk safeguards via a bug bounty program.

Principles

• Proactive red teaming enhances AI safety.
• Universal jailbreaks pose significant risks.

Method

Participants attempt to craft a single prompt that bypasses GPT-5.5's bio safety filters to answer five specific questions without triggering moderation, within a defined timeline and NDA.

In practice

• Apply by June 22, 2026.
• Test GPT-5.5 in Codex Desktop.
• Focus on universal jailbreak prompts.

Topics

• GPT-5.5
• Bio Bug Bounty
• AI Red Teaming
• Biorisk Safety
• Universal Jailbreak

Best for: AI Scientist, CTO, VP of Engineering/Data, AI Security Engineer, Research Scientist

Related on AIssential

• A Red-Team Study of Anthropic Fable 5 & Opus 4.8 Models — Frontier LLMs, despite hardening, remain reliably vulnerable to automated, adaptive jailbreak attacks.
• OpenAI and Anthropic Sign Letter to Prevent AI-Developed Biological Weapons — AI industry leaders are proactively seeking legislation to prevent AI-enabled biological weapons development.
• Introducing the OpenAI Safety Bug Bounty program — OpenAI's new bug bounty targets AI-specific abuse and safety risks beyond traditional security vulnerabilities.
• The White House Wants Anthropic to Block All Jailbreaks. That May Not Be Possible — Preventing all AI "jailbreaks" in advanced models may be technically impossible.
• OpenAI locks GPT-5.5-Cyber behind velvet rope despite slamming Anthropic for doing exactly that — OpenAI is releasing a specialized cyber defense AI model with restricted access, despite previous criticism of similar exclusivity.
• Strengthening societal resilience with Rosalind Biodefense — OpenAI is deploying frontier AI, GPT-Rosalind, to vetted partners for biodefense and pandemic preparedness, emphasizing responsible access and safeguards.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by OpenAI News.