OpenAI strengthens its bug bounty program with rewards up to $172,000

OpenAI significantly raises its bug bounty rewards, with a cap that can reach $172,000 depending on bonuses. The company is also preparing an internal security sprint with the firm Cathedral.

OpenAI has increased its bug bounty program, with rewards that can climb up to $172,000 for certain security discoveries, according to an announcement published on November 3, 2026. Critical vulnerabilities in products like ChatGPT or Codex can now earn external researchers up to $30,000, while exceptional cases can reach $100,000. This overhaul matters because it shows that security is becoming a more aggressive expense item, at a time when models and their tools are multiplying. And, incidentally, it reminds us that there is a more profitable way to break systems than selling them on the black market.

Des primes plus hautes, mais aussi plus fines

  • The new scale is not limited to a brutal increase in amounts.
  • OpenAI adds a five-level loyalty program, as well as temporary campaigns with enhanced rewards during the launch of new features.
  • According to the company, anyone who reported a vulnerability in the last 30 days will retroactively receive the difference between the old and new amounts.
  • This mechanism aims to retain the best bug hunters while directing them towards the most sensitive areas at the moment.

At the same time, OpenAI says it is preparing a "security sprint" with the company Cathedral. The idea is to have internal engineers and external specialists work together to test the robustness of the systems. During the last sprint, 25% of OpenAI's product engineers participated in vulnerability research, according to the company. This type of exercise resembles a technical sparring session: no one wins a medal, but sometimes a more costly future incident is avoided.

Ce que cela dit de la stratégie sécurité d’OpenAI

The new scale places OpenAI in a zone close to Google's general rewards program, according to the cited source, while remaining below the highest bounties offered by Apple or Microsoft for exploits at the operating system level [to be verified]. The comparison is useful because it situates the level of investment sought by OpenAI in the race for security among major platforms. In other words, the company no longer just fixes issues after the fact; it also tries to buy the attention of experts capable of finding vulnerabilities before anyone else.

For AI players, the signal is clear. As products grow in scale, the attack surface also expands, forcing publishers to professionalize their detection and correction systems. In this context, OpenAI relies on a mix of compensation, measured gamification, and direct collaboration with its teams. One simple question remains: if security costs more, how much is a prevented vulnerability worth?

To read