OpenAI's new GPT-6 Cyber finds 81 Firefox flaws and 948 Linux bugs

On Thursday, November 5, OpenAI released GPT-6 Cyber, a model trained to find software flaws and build working attacks on them. The company calls it a "major leap in security research."

The launch comes just 10 days after rival AI lab Anthropic (ANT) publicly filed for its IPO.

Only a small group of vetted customers can use GPT-6 Cyber so far. They come from Daybreak Red, the level of access OpenAI keeps for authorized attack work like penetration testing, where a security team breaks into a system with its owner's permission to find weak spots.

The company says its researchers used the model to uncover 81 new Firefox flaws, all rated high-severity by Mozilla. Mozilla has been quick to ship fixes for many of them.

OpenAI also tested the model against the Linux kernel, the core software of most servers, where it successfully exploited 948 vulnerabilities. That compares with 166 for GPT-5.6 Sol, while 588 of the vulnerabilities were previously unknown, Reuters reported OpenAI saying.

OpenAI rates the model "Critical" in cybersecurity under its Preparedness Framework, the scale it uses to judge how much harm a model could do. Its previous cyber model, GPT-5.6 Cyber, was rated "High."

The company says it sees more attempts to misuse its models, including by hackers and by groups trying to copy them through distillation, which trains a cheaper model on another model's answers. New checks block high-risk work against critical systems such as power plants and water treatment facilities.

Security teams at Cisco (CSCO), Calif, and CodeRabbit already use the model. OpenAI's $1B Daybreak for Frontline Defenders grant program pays for access for teams with small budgets, such as government agencies. All Daybreak Red customers will get GPT-6 Cyber in the coming weeks, OpenAI says.

In June, the U.S. government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 models, which had similar capabilities, for foreign nationals, citing national security authorities. Anthropic said the directive forced it to disable the models for all customers.

Sources & Further Reading

1. "GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents." OpenAI. November 5, 2026.
2. "Anthropic files for IPO, plans Nasdaq listing under "ANT"." StockAnalysis.com. October 26, 2026.
3. "GPT-6 Cyber System Card." OpenAI. November 5, 2026.
4. "OpenAI launches GPT-6 Cyber with 'Critical' cybersecurity rating." Reuters. November 5, 2026.
5. "Statement on the US government directive to suspend access to Fable 5 and Mythos 5." Anthropic. June 12, 2026.