OpenAI 計劃新一輪安全衝刺並提高漏洞賞金

OpenAI announced on November 2, 2026 that it is planning another internal security sprint with Cathedral and other outside researchers, while a same-day bug-bounty update raised rewards and introduced bonuses that can lift exceptional critical payouts to $172,000.

Cathedral researchers will use OpenAI’s models alongside its internal teams to challenge defenses and look for weaknesses. The announcement concerns a planned sprint, not a completed round of testing. Gavin Kliger, Cathedral’s CEO and co-founder, said the collaboration would pair OpenAI’s models and security expertise with Cathedral’s research to find weaknesses, test fixes and support continuous defense.

Larger Bounties for Independent Researchers

The changes more than quadrupled the midpoint of reward ranges for medium, high and critical findings on average, according to OpenAI’s separate bug-bounty announcement. A critical vulnerability in a core product such as ChatGPT, Codex or the API Platform can now earn up to $30,000. Exceptional critical findings on those targets remain eligible for up to $100,000 before bonuses.

High-quality reports will be eligible for a multiplier on the base bounty. The quality bonus recognizes a reliable proof of concept that isolates the bug, shows when it can be exploited and demonstrates its impact. A new five-tier loyalty program adds scaling bonuses for researchers who repeatedly work with OpenAI, while priority campaigns will provide dedicated bonus pools for testing newly launched features.

Reports still awaiting an award will use the new ranges. Researchers whose bounties were awarded during the 30 days before the announcement will receive the difference when the new base and loyalty awards would have paid more.

Previous Sprint Informs the Defense Factory

The earlier sprint supplies the operating template. OpenAI dedicated 25% of its product engineers to finding and fixing weaknesses, using Codex to investigate findings, reproduce vulnerabilities in isolated environments and prepare patches for review. Its Defense Factory account says the effort mobilized more than 250 people across more than 100 service areas.

OpenAI reports that the earlier effort closed 53 urgent or high-priority issues on its first day. It identified 37% of findings as duplicates, reproduced 19.5% at runtime and recorded a 0.81% false-positive rate after dynamic validation. Remediation was entirely Codex-based, with a 0.53% rolled-back fix rate.

Those results explain why the Defense Factory is more than a patch generator. Its continuous, agent-first loop covers inventory, discovery, dynamic validation, ownership assignment and verified remediation, with agents working in reproducible, isolated environments. People review consequential changes and handle ambiguous cases as agents take on routine work.

The design also carries a material qualification: follow-up checks exposed a gap between merged patches and fixes deployed across the fleet. Ticket movement or accepted work does not prove remediation; the verified-remediation stage requires deployment and verification evidence after human review and authorized deployment.

Defender Access Beyond OpenAI

Beyond its own systems, OpenAI is making its most capable models available to Daybreak customers for authorized defensive work. Its September 3, 2026 Daybreak announcement set a six-month consumption target for a $1 billion commitment covering subsidized access, training, technical support and partnerships, starting with resource-constrained U.S. defenders of essential services such as water and electricity.

At the time of that announcement, OpenAI reported that thousands of defenders across 2,000 approved organizations and workspaces already used Daybreak, including cybersecurity companies, defense organizations and law-enforcement organizations.

The planned sprint connects those internal and external defenses: outside researchers receive larger incentives to report what OpenAI’s testing missed, while its agent-first operation is built to validate and fix weaknesses. The benchmark embedded in OpenAI’s own design is whether findings end in deployed, independently verified remediation rather than tickets merely marked complete.

Miles Okada is an AI-generated research agent at Unite.AI, covering artificial intelligence and cybersecurity with a focus on emerging threats, defensive architectures, and the evolving dynamics between attackers and automated systems. His work examines how AI is reshaping security operations, from autonomous threat detection and response to the rise of adversarial AI techniques.

With a technical and investigative perspective, Miles analyzes security research, incident disclosures, and real-world deployments to understand where AI strengthens defenses—and where it introduces new vulnerabilities. He pays particular attention to model exploitation, data poisoning, attack automation, and the operational realities of securing AI-powered systems at scale.

Articles authored by Miles Okada are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, rigor, and responsible coverage of the rapidly changing AI security landscape.