explainx / blog
OpenAI July 9, 2026 doubles Bio Bug Bounty to $50,000 for a universal jailbreak against GPT-5.6 and GPT-5.5 biosafety challenges. Private program, NDA, scope until July 27 for 5.5. vs Fable red-team jokes and payout skepticism.
Jul 9, 2026
OpenAI launches GPT-5.6 publicly July 9; Fable 5 live globally since July 1. Sol Ultra leads Terminal-Bench at 91.9%; Fable leads SWE-Bench Pro at 80.3%. Terra matches Fable on terminal work at half the price — tier-by-tier guide.
Jul 1, 2026
GeneBench-Pro measures whether AI agents can do judgment-heavy computational biology — not just run a pipeline on clean data. GPT-5.6 Sol leads at ~29–32% pass rate on problems that cost human experts thousands of dollars each. Ten questions open-sourced on Hugging Face; 50-question subset heading to Artificial Analysis.
Jun 28, 2026
Dario Amodei helped decide not to release GPT-2 in full in 2019 because OpenAI feared misuse. Seven years later, he runs Anthropic, keeps Claude completely closed, and finds himself at the epicentre of the most heated open-source AI debate the industry has ever seen.
On July 9, 2026 — the same day GPT-5.6 Sol, Terra, and Luna rolled out — @OpenAI posted OpenAI Bio Bug Bounty: the GPT-5.5 Bio Bug Bounty becomes an ongoing private program, rewards double to $50,000, and GPT-5.6 enters scope from launch day.
The hunt: a universal jailbreak — one technique that defeats OpenAI's predefined biosafety challenge against harmful biology outputs (pathogen-style instructions), not a single lucky prompt on one session.
| Field | Detail |
|---|---|
| Program | OpenAI Bio Bounty Program (evolved from GPT-5.5 Bio Bug Bounty) |
| Top prize | $50,000 (was $25,000) for universal jailbreak |
| In scope | GPT-5.6 from day one · GPT-5.5 through July 27, 2026 |
| After Jul 27 | GPT-5.6 only (unless OpenAI updates scope) |
| Partial wins | Smaller awards at OpenAI discretion |
| Access | Private · rolling application · NDA · ChatGPT account required |
| Reapply? | Past GPT-5.5 applicants do not need to reapply |
From the official post:
The program will remain focused on universal jailbreaks that can defeat our predefined biosafety challenge against OpenAI's frontier models, starting with GPT-5.6 and going forward.
Universal is the loaded word. X reactions split:
explainx.ai read: This is safety marketing + real testing in one package — parallel to GPT-5.6's government-gated preview and Washington's interest in frontier biology risk, not a consumer feature.
| Date | Scope |
|---|---|
| Before July 9 | GPT-5.5 Bio Bug Bounty at $25K |
| July 9, 2026 | Program renamed · $50K · GPT-5.6 added |
| Through July 27, 2026 | GPT-5.5 still testable under original scope |
| After July 27, 2026 | GPT-5.6 primary in-scope model |
OpenAI will communicate future scope changes to onboarded researchers.
Related programs: Safety Bug Bounty and Security Bug Bounty for non-bio classes.
July 2026 stack:
| Layer | OpenAI signal |
|---|---|
| Capability | GPT-5.6 Sol biology + GeneBench-Pro |
| Cyber | Sol does not cross Cyber Critical per preview system card |
| Bio safeguards | Layered refusals + classifiers + $50K external red team |
| Policy | Same week as Fable export-control discourse — different mechanism (bounty vs ban) |
Anthropic's parallel story: classifier-layer Fable for public access vs Mythos for vetted cyber/bio. OpenAI's parallel: pay researchers to break bio guards on GPT-5.6 while shipping it broadly.
Official: OpenAI Bio Bug Bounty · Apply
Bounty amounts, scope dates, and program rules reflect OpenAI's July 9, 2026 announcement. Private program details may change — verify on openai.com before applying.
