OpenAI raises bug bounty rewards, but evidence still sets the bar

OpenAI says it has more than quadrupled the midpoint of its reward ranges for medium, high and critical security findings on average. The November 2 update also says a critical vulnerability in core products such as ChatGPT, Codex or the API Platform can earn up to $30,000, with exceptional findings eligible for more through bonuses.

The increase signals that OpenAI wants more independent researchers to look for security flaws its internal testing missed. But the headline maximum is not a typical payout or a promise attached to every report. Eligibility, impact, scope and report quality remain central, and the program’s current terms are the source of truth for any researcher considering work.

OpenAI says exceptional critical findings on core targets remain eligible for up to $100,000, and up to $172,000 with bonuses. It also describes new incentives for higher-quality vulnerability reports. These figures describe ceilings under stated conditions; they should not be read as a guaranteed rate or as a reason to test outside the published rules.

A bounty rewards a demonstrated security failure

A useful report connects an in-scope weakness to a concrete impact, explains the conditions required to reproduce it, and gives the security team enough evidence to validate and fix it. That is different from a product complaint, a speculative risk, or a general model behavior that does not establish a security vulnerability under program policy.

Before testing, researchers should read the current Bugcrowd scope and rules, use accounts and data they control, and avoid actions that could affect other users or service availability. If the program disallows a test or the boundary is unclear, stop and ask the program through its designated channel. Safe testing is part of the quality of the finding, not an administrative afterthought.

OpenAI’s reward changes also highlight a broader reality: security research is a collaboration. A report that includes a minimal reproduction, impact assessment and responsible handling can reduce triage time and help teams distinguish a real vulnerability from an ambiguous observation. That work has value even when a finding does not qualify for the largest amount.

What researchers should do next

Treat the announcement as a reason to re-read the policy, not to assume that all categories or targets have changed. Check exact scope, exclusions, submission requirements, and how severity is assessed. Keep evidence limited to what demonstrates the issue, redact unrelated personal information, and do not publish details before the program’s coordinated disclosure process allows it.

For organizations operating bounty programs, the lesson is similar: reward structure can attract attention, but a clear scope and predictable triage process shape the quality of what arrives. Higher payouts alone cannot substitute for a responsive security team and a path from validated report to shipped fix.

OpenAI’s update raises the potential compensation for serious findings and explicitly adds quality incentives. Researchers should still optimize for a precise, safe and reproducible report. The bounty is the reward mechanism; the evidence is what makes a finding actionable. For more on security failures in everyday software supply chains and operations, read our analysis of the LiteLLM package compromise and email security exceptions.