November 3, 2026, (Inside AI) — OpenAI will launch its next internal security sprint, bringing outside researchers and security partners together with its own teams to probe its systems using its own AI models, the company announced. The effort includes a partnership with security firm Cathedral and a major expansion of its bug bounty payouts, with rewards for critical findings more than quadrupled.
The move reflects a widening bet that frontier AI models can serve as defensive tools, not just attack surfaces. OpenAI says traditional cyber defenses alone can no longer keep pace with threats that emerge faster than human review cycles can absorb.
OpenAI is positioning the sprint as a continuous loop: use models to find weaknesses, reproduce them in isolated environments, prepare patches, and repeat. The company said its previous sprint dedicated 25% of its product engineers to finding and fixing weaknesses, with engineers using Codex to investigate findings and draft patches for review.
The new sprint adds external pressure. Cathedral's researchers will work alongside OpenAI teams to challenge defenses, a structure designed to bring outside assumptions into a process that could otherwise become insular.
"We want researchers to challenge our assumptions and help strengthen both our systems and the methods we use to defend them." (OpenAI, in a company statement)
The bounty expansion is the most concrete incentive change. More than quadrupling rewards for critical findings signals that OpenAI wants deeper, more expensive vulnerability research, not just surface-level reports. The company did not disclose specific dollar figures for the new reward tiers.
OpenAI also committed $1 billion in subsidized access over six months for frontline defenders protecting essential services and digital infrastructure, and said it will make its most capable models available to Daybreak customers.
Why AI Defenders Need Human Adversaries
The partnership model matters because AI-driven vulnerability discovery has a known failure mode: models can generate plausible findings that do not hold up under scrutiny. Human researchers with different expertise can filter noise and probe assumptions that automated systems inherit from their training data.
OpenAI's decision to publish its architecture and lessons from the last sprint, alongside a blueprint for cyber defenders across industry and government, suggests the company wants the approach adopted beyond its own walls. That is a shift from the usual pattern, where security methods stay proprietary.
The competitive context is sharp. Anthropic, Google, and Microsoft have all invested in AI-assisted security research, and governments are pressing critical infrastructure operators to harden systems against AI-enabled attacks. OpenAI's $1 billion subsidy is a market-making move as much as a security one.
OpenAI's own framing is blunt about the stakes. It said recent incidents across the industry show how quickly threats can emerge, and that keeping the defender's window open requires organizations to use frontier cyber capabilities to harden their systems.
The company said it will keep advancing its models, using them to strengthen its own systems, and expanding access for trusted defenders. It added that keeping the defender's window open will take action across the industry, and that work needs to start now.
Tobias is less interested in what a robot or a prompt can do today than in what it means for tomorrow. He writes about prompt engineering and robotics for readers who want context, not just specs.
