How Much AI Labs Pay for Security Bugs and AI Jailbreaks

OpenAI raised the rewards in its Security Bug Bounty on Nov 2. A critical vulnerability in a core product such as ChatGPT, Codex or the API Platform now earns up to $30,000, and exceptional critical findings on those products "remain eligible for up to $100,000, and now up to $172,000 with bonuses," the company said in a blog post. The labs' own bounty pages put that figure below Meta's top of $300,000, and every jailbreak bounty they publish pays $50,000 or less.

OpenAI's top payout since 2023

OpenAI launched its bug bounty on Bugcrowd on Apr 11, 2023, with rewards "from $200 for low-severity findings to up to $20,000 for exceptional discoveries." On Mar 26, 2025 it raised the maximum for exceptional and differentiated critical findings to $100,000. The Nov 2 raise keeps that $100,000 and adds bonus multipliers for report quality, a five-tier loyalty program and short-term priority campaigns on new features. OpenAI's own example stacks all three, a quality bonus, a priority campaign and its 31337 loyalty tier, on $100,000 to reach $171,925, which the post rounds to $172,000.

Security bugs pay the most, AI agents included

Meta's payout guidelines list up to $300,000 for mobile remote code execution in apps such as WhatsApp and Instagram, and for WhatsApp Private Processing. Muse, which Meta describes as "an AI product from Meta" that "lets people create their own personalized AI agent," pays up to $300,000 for a compromise of Meta production and users beyond Muse, and up to $130,000 for a compromise of a specific user's Muse agent. Meta AI issues such as being able to access private user content pay up to $130,000.

Google's Vulnerability Reward Program pays a base reward of up to $101,010 for remote code execution, before bonuses for report quality and novelty. Its AI Vulnerability Reward Program, which Google launched on Oct 6, 2025, pays a top base reward of $20,000 for rogue actions, attacks that change a victim's account or data, in its flagship AI products: Search, the Gemini apps and Workspace's core apps such as Gmail and Docs. xAI's security.txt lists $7,500 to $20,000 for critical bugs.

Jailbreak bounties stop at $50,000

Anthropic offered up to $15,000 for universal jailbreaks on Aug 8, 2024, and up to $25,000 in a program launched on May 14, 2025 that ran through May 18. Its bounty page, dated Sep 8, 2026, pays up to $35,000 per novel universal jailbreak. OpenAI raised its bio bounty for a universal jailbreak from $25,000 to $50,000 on Jul 9, 2026. Both are invite-only programs whose participants sign a non-disclosure agreement.

The other programs leave jailbreaks out. Google's AI program lists them among the issues that don't qualify, Meta's AI guidelines treat them as payout-ineligible unless chained with another eligible risk, and OpenAI's Safety Bug Bounty, launched on Mar 25, 2026, says jailbreaks are out of its scope.

Where no amount is published

Anthropic's bounty page sends security vulnerabilities to its Responsible Disclosure Policy, which names no reward. xAI asks for model safety issues, jailbreaking among them, to be emailed to its AI safety team, with no reward listed. Meta's guidelines, Google's rules and xAI's security.txt carry no date, so their figures are as read on Nov 4, 2026.