OpenAI doubles bio-jailbreak bounty to $50,000 as GPT-5.6 lands

The News

OpenAI has widened its Bio Bug Bounty, inviting outside researchers to hunt for a single universal jailbreak that would strip the biosafety guardrails from its latest models. The top reward now stands at $50,000, doubled from the $25,000 the company previously offered.

The programme covers two models: the newly released GPT-5.6 and its predecessor GPT-5.5. Researchers have until 27 July 2026 to probe GPT-5.5; after that date, only GPT-5.6 remains in scope. Smaller partial-win payments are available at OpenAI's discretion for submissions that fall short of a full universal break.

Participation is deliberately gated. Applicants apply on a rolling basis through a Google Form, must hold an existing ChatGPT account, and are required to sign a non-disclosure agreement while sharing their name, affiliation and relevant experience. OpenAI frames the exercise as part of its wider push to "strengthen our safeguards for advanced AI capabilities in biology". Past GPT-5.5 applicants need not reapply.

Why It Matters

A bounty aimed specifically at biological misuse is a different animal from the usual security payout. Rather than rewarding a leaked API key or a cross-site scripting flaw, OpenAI is paying people to prove that its refusal training can be coaxed into producing dangerous biology content. Doubling the headline figure signals both confidence and unease: confidence that the guardrails will hold, and worry about what a single reusable exploit could unlock.

The move tracks the trajectory of the wider AI-safety field. When the big labs first opened red-teaming programmes around frontier models in 2023, biosecurity sat near the top of the risk register alongside cyber weapons. Three years on, that threat model has hardened into a standing, cash-backed contest. Scoping two model generations at once also hints at release cadence: GPT-5.6 arriving while GPT-5.5 is still fresh suggests OpenAI is shipping iterative upgrades far faster than the multi-year gaps of the GPT-3 to GPT-4 era.

Indian Angle

For India, the interesting question is who is likely to claim that $50,000, roughly Rs 42 lakh at current rates. Indian researchers routinely top global bug-bounty leaderboards on platforms such as HackerOne and Bugcrowd, and the country supplies one of the largest pools of independent security testers anywhere. A biosafety bounty with a five-figure dollar payout is squarely the kind of contest where Bengaluru and Hyderabad hackers have historically punched above their weight.

There is a policy dimension too. India's own model builders, Sarvam and Ola-backed Krutrim among them, are training frontier-class systems without anything like OpenAI's public biosecurity apparatus. As the IndiaAI Mission and MeitY's nascent AI Safety Institute shape domestic guardrail norms, OpenAI's paid disclosure model offers a ready template for how Indian labs might invite scrutiny rather than wait for regulators to impose it.

Finally, cost. A $50,000 reward is trivial for a firm of OpenAI's scale but transformative for an individual Indian researcher, where it can equal several years of a mid-level engineer's salary. That asymmetry is precisely what makes global bounties such an efficient way to tap Indian talent without formally hiring it.

FAQ

When is the deadline for GPT-5.5?

Researchers have until 27 July 2026 to test GPT-5.5. After that date, only the newer GPT-5.6 stays eligible for submissions under the programme, so anyone targeting the older model needs to move quickly.

How much can a researcher earn?

A confirmed universal jailbreak pays $50,000 for either GPT-5.6 or GPT-5.5, up from the previous $25,000. Partial or lesser findings may earn smaller amounts, awarded entirely at OpenAI's discretion.

How do you take part?

Apply through OpenAI's rolling Google Form. You need an existing ChatGPT account, must sign a non-disclosure agreement, and should provide your name, affiliation and experience. Researchers who applied for the earlier GPT-5.5 round do not need to reapply.

Where can I read the original announcement?

OpenAI published the full details on its official blog, which is linked in the attribution paragraph directly below this section.

This story was reported by OpenAI. Read the full original coverage at OpenAI.