OpenAI Quadruples Bug Bounty Rewards, Partners With Cathedral on Security Hardening Sprint
OpenAI is launching a new security sprint with external researchers and quadrupling critical vulnerability rewards as part of an AI-driven defense initiative.
Last verified:
OpenAI Escalates AI-Powered Security Testing With Expanded Bug Bounty Program
OpenAI is launching a new internal security sprint designed to stress-test its platform defenses by partnering with external researchers and deploying its own frontier models to uncover weaknesses. According to the OpenAI Blog, the initiative brings Cathedral—a security research firm—together with OpenAI’s internal teams to probe the company’s systems using offensive security techniques and differentiated vulnerability discovery methods.
The announcement signals a shift in OpenAI’s defensive posture: rather than waiting for attackers or passive disclosure channels to surface flaws, the company is embedding continuous vulnerability research into its product engineering cycle. The expanded bug bounty program more than quadruples reward payouts for critical findings, signaling that OpenAI views external researcher participation as a core pillar of its security architecture.
AI Models as Active Defenders
OpenAI’s rationale hinges on the observation that traditional reactive cyber defense—firewalls, intrusion detection, patching cadences—no longer closes the gap between threat emergence and defensive response. The company’s previous security sprint, which allocated 25% of product engineering capacity to vulnerability hunting, demonstrated that AI models like Codex could accelerate the vulnerability investigation and patch preparation cycle by automating reproduction of flaws in isolated sandbox environments.
This sprint operationalizes that lesson at scale. By deploying frontier-capability models to actively hunt for weaknesses in OpenAI’s own infrastructure, the company aims to compress the defender’s time-to-patch window—the interval during which attackers can exploit unpatched vulnerabilities. According to the OpenAI Blog, the approach reflects industry-wide recognition that recent security incidents have shown how rapidly threats propagate and how critical it is to establish continuous defensive feedback loops.
Subsidy Strategy for Critical Infrastructure Defenders
OpenAI is extending access to its security-focused model capabilities beyond its own walls. The company committed $1 billion in subsidized model access over six months through its Daybreak program, targeting defenders who protect essential services and digital infrastructure. This dual-track approach—hardening OpenAI’s own systems while arming trusted external defenders with the same frontier capabilities—attempts to shift the industry-wide defender-to-attacker ratio by democratizing access to AI-powered security tooling.
The company plans to release its defensive architecture and lessons learned from the previous sprint, positioning itself as both a security consumer and a security technology provider for the broader ecosystem.
Why This Matters
OpenAI’s announcement reflects a strategic bet that AI models will become the primary lever for closing security gaps faster than attackers can exploit them. For security teams evaluating frontier LLM adoption, this signals OpenAI’s confidence in using its own models for adversarial red-teaming—a capability that organizations without in-house model development capacity may lack. The quadrupled bug bounty program also raises the bar for how security researchers are compensated; teams managing vulnerability disclosure programs may face pressure to match or exceed these payouts. For defenders in critical infrastructure sectors, the $1 billion Daybreak subsidy makes frontier model access economically viable for security automation at scale—though the six-month commitment window suggests the program is time-bound and subject to future negotiation.
Frequently Asked Questions
What is the Cathedral partnership?
Cathedral is a security research organization whose experts will collaborate with OpenAI's teams to test defenses and identify vulnerabilities using OpenAI's models.
How much will OpenAI spend on security researcher subsidies?
OpenAI committed $1 billion in subsidized model access over six months for defenders protecting critical infrastructure and digital services, through its Daybreak program.
What did the previous security sprint accomplish?
OpenAI dedicated 25% of product engineers to vulnerability research, using Codex to investigate findings and prepare patches in isolated test environments.