SafetyRisk, alignment & guardrails
OpenAI raises its top bug bounty to about $172,000 and adds bonuses for report quality, loyalty and priority campaigns
OpenAI has raised Security Bug Bounty rewards for ChatGPT, Codex and its API. Exceptional critical findings can now reach about $172,000 once new bonuses for report quality, loyalty and priority campaigns stack, up from a $100,000 cap set in 2025.
From the OpenAI News article of November 2, 2026 and cited sources: The new structure pays up to $30,000 base for critical bugs and up to $100,000 base for exceptional critical findings. [1] [2] [3]
SOURCES[1] OpenAI: Rewarding defenders with bigger bug bounties openai.com[2] BleepingComputer: OpenAI now pays researchers $100,000 for… bleepingcomputer.com[3] ShiftDelete: OpenAI Bug Bounty Now Pays Up to $172,000 shiftdelete.net
AT Pulse research deskPublished · Updated · 4 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- The new structure pays up to $30,000 base for critical bugs and up to $100,000 base for exceptional critical findings. [1] [2] [3]
- Original source
- OpenAI News article, company source
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 2, 2026 article published
- Covered
- Nov 3 edition
- Evidence
- 6 sources across 6 sites
- 1OpenAI: Rewarding defenders with bigger bug bountiesopenai.com
- 2BleepingComputer: OpenAI now pays researchers $100,000 for…bleepingcomputer.com
- 3ShiftDelete: OpenAI Bug Bounty Now Pays Up to $172,000shiftdelete.net
- 4Interesting Engineering: OpenAI expands security testinginterestingengineering.com
- 5SuperPower Daily: OpenAI raises critical bug bounties more…superpowerdaily.com
- 6NewsBytes: OpenAI partners with Cathedral Systemsnewsbytesapp.com
Reach: not measured yet
The new structure pays up to $30,000 base for critical bugs and up to $100,000 base for exceptional critical findings. Stacked bonuses for report quality, loyalty and priority campaigns lift the maximum to about $172,000. The program launched in 2023 with a $20,000 cap, which rose to $100,000 in March 2025. [1] [2] [3]
The headline claim that rewards 'more than quadruple' refers to the average midpoint of the medium, high and critical ranges, not the ceiling. The top cap rises only about 1.7 times. Examples such as critical account takeover going from $6,500 to $51,577 assume the maximum bonuses, so they are not typical payouts. [3] [5]
The quality bonus applies when a report includes a reliable proof of concept that isolates the bug and shows its impact. Clear, reproducible reports now earn more money, and the loyalty tiers favor researchers who keep working on the same program. [3]
Bigger rewards weighted toward bonuses should bring more outside testing of widely used AI products and push rival vendors to raise their bounties. Payout tables show incentives, not results.
Read the full assessment
No reviewed source reports disclosure counts, fix times or how often top awards are paid.
Executive brief
OpenAI now says one exceptional critical bug in ChatGPT, Codex or the API Platform can earn about $172,000 once bonuses stack. In 2023 the program launched with a $20,000 cap. The company announced the changes on 2 November 2026. It says it is raising base rewards and adding bonuses for report quality, loyalty and priority campaigns (OpenAI). The bounty news came alongside an internal "security sprint" with the outside firm Cathedral and a subsidized model-access program for cyber defenders. No independent data on actual payouts is available.
Read the full section
OpenAI now says one exceptional critical bug in ChatGPT, Codex or the API Platform can earn about $172,000 once bonuses stack. In 2023 the program launched with a $20,000 cap. The company announced the changes on 2 November 2026. It says it is raising base rewards and adding bonuses for report quality, loyalty and priority campaigns (OpenAI). The bounty news came alongside an internal "security sprint" with the outside firm Cathedral and a subsidized model-access program for cyber defenders. All the numbers come from OpenAI. No independent data on actual payouts is available.
What changed and event timeline
Bounty program launches
OpenAI opened its program on Bugcrowd, paying up to $20,000 for vulnerabilities in its products ().
Top payout rises to $100,000
The cap for "exceptional and differentiated" critical findings went up fivefold. A time-limited bonus doubled IDOR (access-control) rewards to $400–$13,000 until 30 April 2025 ().
Bigger bounties and three bonus types
OpenAI raised Security Bug Bounty rewards and added report-quality, loyalty and priority-campaign bonuses (;).
Security sprint with Cathedral
Cathedral researchers will test OpenAI's defenses and fixes. In an earlier sprint, 25% of OpenAI's product engineers used Codex to reproduce vulnerabilities. No start date has been given ().
20–21 November 2026 (scheduled): PROMPT2PWN in Oakland
Only one reviewed source mentions this event, so treat it as unconfirmed ().
Capabilities and access
- Products in scope for top rewards: ChatGPT, Codex and the API Platform (Help Net Security).
- Base pay: critical bugs up to $30,000. Exceptional critical findings up to $100,000 base. Total with bonuses up to about $172,000 (ShiftDelete).
- Loyalty: five tiers, with larger bonuses at higher tiers.
Read the full section
- Products in scope for top rewards: ChatGPT, Codex and the API Platform (Help Net Security).
- Base pay: critical bugs up to $30,000; exceptional critical findings up to $100,000 base; total with bonuses up to about $172,000 (ShiftDelete).
- Loyalty: five tiers, with larger bonuses at higher tiers.
- Back pay: reports not yet paid get the new ranges. Bounties paid in the 30 days before the announcement get topped up to the new level (Help Net Security).
Technical analysis for researchers and developers
- The quality multiplier applies to reports with a reliable proof of concept that isolates the bug and shows its impact (ShiftDelete).
- Priority campaigns are short bonus pools for newly launched features (Help Net Security), so researchers are pushed toward the newest code.
- OpenAI's "Defense Factory" team reportedly uses the company's own models to find vulnerabilities and test fixes (Help Net Security).
Read the full section
- Proof of concept now pays directly. The quality multiplier applies to reports with a reliable proof of concept that isolates the bug and shows its impact (ShiftDelete). A clean, reproducible report becomes part of the payout, not just good manners.
- Campaigns target new features. Priority campaigns are short bonus pools for newly launched features (Help Net Security), so researchers are pushed toward the newest code.
- Model-assisted triage. OpenAI's "Defense Factory" team reportedly uses the company's own models to find vulnerabilities and test fixes (Help Net Security). No method or evaluation for it has been published.
Claims and evidence
- "More than quadruple": OpenAI says the average midpoint of medium, high and critical reward ranges more than quadruples (ShiftDelete).
- Example before/after figures: critical account takeover, $6,500 → $51,577; a medium bug, $750 → $2,415. Both are OpenAI's top-bonus examples, not typical payouts (ShiftDelete).
- Corroboration: every outlet reviewed restates OpenAI's announcement. No independent data exists on how often top payouts are made.
Read the full section
- "More than quadruple": OpenAI says the average midpoint of medium, high and critical reward ranges more than quadruples (ShiftDelete). That is not the same as quadrupling the maximum: the top cap rises from $100,000 to about $172,000, roughly 1.7×.
- Example before/after figures: critical account takeover, $6,500 → $51,577; a medium bug, $750 → $2,415. Both are OpenAI's top-bonus examples, not typical payouts (ShiftDelete).
- Corroboration: every outlet reviewed restates OpenAI's announcement. No independent data exists on how often top payouts are made.
Context and prior work
- The 2025 increase raised the cap from $20,000 to $100,000 and added time-limited IDOR bonuses (BleepingComputer).
- The 2026 package also includes $1 billion in subsidized model access over six months for defenders of critical infrastructure (Interesting Engineering).
- OpenAI says it will share what it learned from the earlier sprint with industry and government defenders (SuperPower Daily).
Read the full section
- The 2025 increase raised the cap from $20,000 to $100,000 and added time-limited IDOR bonuses (BleepingComputer).
- The 2026 package also includes $1 billion in subsidized model access over six months for defenders of critical infrastructure (Interesting Engineering). ShiftDelete calls this program "Daybreak" (ShiftDelete).
- OpenAI says it will share what it learned from the earlier sprint with industry and government defenders (SuperPower Daily).
Limitations, safety and contested findings
- Figures are reported inconsistently. Headlines say "$172,000"; ShiftDelete gives $171,925.
- It appears as "Cathedral" in some outlets and "Cathedral Systems" in others (NewsBytes).
- The reviewed sources don't explain how the multipliers or loyalty-tier thresholds are calculated.
Read the full section
- Figures are reported inconsistently. Headlines say "$172,000"; ShiftDelete gives $171,925. Help Net Security calls $100,000 a base amount for exceptional findings, while other coverage leads with the $30,000 critical base.
- The partner's name varies. It appears as "Cathedral" in some outlets and "Cathedral Systems" in others (NewsBytes).
- Missing details. The reviewed sources don't explain how the multipliers or loyalty-tier thresholds are calculated. The quote from Cathedral's CEO is praise from a commercial partner (Interesting Engineering).
Business and practitioner implications
- For security researchers: OpenAI's core products now pay much more than before.
- For other AI vendors: the bar for bounties is higher. Expect pressure to match on top-end pay and on campaigns tied to product launches.
- For enterprise buyers: more outside testing of ChatGPT, Codex and the API Platform helps. Still, payout tables don't show results.
Read the full section
- For security researchers: OpenAI's core products now pay much more than before. Repeat work and well-documented proof-of-concept reports earn the most, which favors specialists who stick with one program.
- For other AI vendors: the bar for bounties is higher. Expect pressure to match on top-end pay and on campaigns tied to product launches.
- For enterprise buyers: more outside testing of ChatGPT, Codex and the API Platform helps. Still, payout tables don't show results. Look for disclosure statistics and fix times, which no reviewed source reports.