Find a GPT-5 jailbreak and win $25,000 from OpenAI

OpenAI’s selective bug bounty, limited to vetted red-teamers, includes a $10,000 prize for solving all jailbreak challenges via multiple prompts, smaller payouts for partial progress, with formal testing kicking off September 9, 2025

OpenAI has rolled out a new bug bounty initiative for its latest large language model, GPT-5, offering security researchers and red-teamers the chance to test its resilience against jailbreak attempts. The program’s headline reward stands at $25,000 for the first participant who successfully develops a universal jailbreak prompt capable of bypassing GPT-5’s moderation systems and forcing it to answer ten sensitive bio- and chemistry-related safety questions.

Alongside the top payout, OpenAI has announced a $10,000 prize for the first team to solve all ten questions using multiple jailbreak prompts. The company added that smaller rewards may also be granted for partial breakthroughs, depending on the discretion of its review panel.

Participation, however, remains highly selective. Applications are currently invite-only, restricted to vetted biosecurity experts and approved red-teamers. Once onboarded, participants will gain access to OpenAI’s dedicated bio bug bounty platform. Testing is scheduled to begin on September 9, 2025. To ensure safe handling of discoveries, all submissions and related communications will be covered under strict non-disclosure agreements.

Looking ahead to GPT-6

Even as GPT-5 undergoes rigorous testing, OpenAI has its sights set on the next chapter. CEO Sam Altman, speaking in San Francisco, hinted that GPT-6 could be released sooner than the gap observed between GPT-4 and GPT-5. The upcoming model, he suggested, may redefine user interaction by shifting from reactive responses to adaptive, personalized engagement.

Altman emphasized memory as a cornerstone feature, enabling users to create tailored AI assistants that align with individual routines and preferences. This personalization, he argued, will be key to making ChatGPT “truly personal.”

For now, OpenAI’s immediate priority remains on GPT-5 and whether external experts can challenge its guardrails. The outcome of the bug bounty program could play a critical role in strengthening safeguards around advanced AI systems before the company advances to its next generation.