OpenAI’s GPT-6 Cyber Found 81 Firefox Flaws and 948 Linux Exploits

OpenAI's newest security model is rated "Critical" under its own risk framework. Here is what it found, who can use it, and what it means for Firefox and Linux users in Canada and the U.S.

Jach Hoggen
15 Min Read
Server racks in a data centre. Photo: Carl Lender / Wikimedia Commons, CC BY 2.0
Highlights
  • OpenAI announced GPT-6 Cyber on November 5, 2026, a security model it says found 81 previously unknown high-severity vulnerabilities in Firefox.
  • In a mid-October Linux kernel run, OpenAI's pipeline reproduced 1,724 vulnerabilities and produced 948 working root exploits at an API cost of about $1.2 million.
  • OpenAI rates the model "Critical" for cybersecurity under its Preparedness Framework, up from "High" for GPT-5.6 Cyber.
  • Access is limited to approved customers in OpenAI's Daybreak Red program, with IP allowlisting required by default.
  • Mozilla's Firefox 159 advisory lists 167 CVEs, 41 of them credited to OpenAI researchers or OpenAI Codex Security; update Firefox and Linux systems promptly.

OpenAI says its new GPT-6 Cyber model found 81 previously unknown high-severity security flaws in Firefox and built 948 working exploits for Linux kernel bugs. The model is not public: it is limited to vetted security teams, and OpenAI itself classifies its hacking ability as “Critical” risk. Source: OpenAI.

The figures come from OpenAI’s own announcement and research post, not independent testing. Mozilla’s published advisories and reporting by BleepingComputer support the broad picture of AI-driven bug hunting at a new scale. Source: Mozilla. Source: BleepingComputer.

Here is what the model did, how the Linux project worked, what the safeguards are, and what ordinary Firefox and Linux users in Canada and the U.S. should do now.

What OpenAI Announced on November 5

GPT-6 Cyber is a version of OpenAI’s models tuned for security research: finding vulnerabilities, proving they can be exploited, and drafting patches. OpenAI describes it as more capable than GPT-5.6 Cyber, its previous security model, and better at staying inside the limits a researcher sets. Source: OpenAI.

The company says security teams at Cisco, Calif and CodeRabbit already use it for vulnerability research and code review. OpenAI also says it used the model to harden its own infrastructure. Source: OpenAI.

The model is available through OpenAI’s API and its Codex coding tool, but only to a select group of approved Daybreak Red customers. OpenAI says it will expand access to all Daybreak Red customers soon. Daybreak is OpenAI’s security program: Daybreak Blue serves verified defenders, while Daybreak Red allows deeper exploit development and authorized testing. Source: OpenAI.

OpenAI framed the release as a race. The company wrote that there is “a limited window to use frontier models to secure software,” before similar capabilities spread more widely. Source: OpenAI. Source: BleepingComputer.

The Firefox Findings: 81 New High-Severity Bugs

During a vulnerability research campaign, OpenAI says GPT-6 Cyber produced evidence and reproducible tests for 81 new vulnerabilities that Mozilla classified as high severity. It also showed that 30 already-known issues were more severe than first reported. Source: OpenAI.

OpenAI says it sent the findings to Mozilla with proposed patches, and that Mozilla began landing fixes shortly after. The company says fixes for many, not all, of the Firefox flaws have shipped. Source: OpenAI.

Mozilla’s own records show how large the recent wave is. Its security advisory for Firefox 159, dated October 27, 2026, lists 167 CVEs: 108 high, 52 moderate and 7 low. Of those, 41 high-severity entries credit OpenAI, either “Amy Burnett of OpenAI” (7) or “OpenAI Codex Security” (34). Source: Mozilla.

The Mozilla advisory does not name GPT-6 Cyber, so readers cannot match individual CVEs to the 81 bugs from public records alone. Use-after-free bugs, where a program keeps using memory it has already released, are the most common type listed. Source: Mozilla.

Mozilla also changed how it reports. The advisory says it no longer rolls all internally found memory safety bugs into a single CVE, which partly explains why the count per release looks so high. Source: Mozilla.

GPT-6 Cyber is a world-class model for cyberdefense.

Brian Grinstead, Distinguished Engineer, Firefox, as quoted by OpenAI

How the Linux Kernel Project Worked

OpenAI described the Linux work in a separate research post published November 4. It built an automated pipeline that scans the kernel, reproduces suspected bugs in isolated virtual machines, tests whether they can be exploited, writes candidate patches and then hands everything to a human researcher. Source: OpenAI.

The Linux kernel is the core of the operating system behind Android phones, cloud servers and supercomputers. OpenAI notes it has more than 40 million lines of code. Source: OpenAI.

Scout agents first picked 5,034 audit targets, about 2.8 million lines, or roughly 7% of the kernel. Analyst agents then produced 2,308 candidate findings, and validation agents reproduced 1,724 of them, about 74.7%. Source: OpenAI.

The result was 948 working local privilege escalation exploits, meaning an ordinary user account on the machine could gain full administrator (root) control. The run also produced proof-of-concept code for 291 information-leak bugs. Source: OpenAI.

OpenAI says only 3.1% of candidates reflected incorrect analysis by the model. Most of the rest needed elevated privileges or could not be reproduced within the time budget. Source: OpenAI.

  • Discovery: agents choose reachable, security-relevant code and scan it.
  • Validation: each candidate is run in a sandboxed kernel with memory-error detection.
  • Exploitability: agents test whether a bug leaks data or grants root.
  • Patching and review: candidate fixes go to a human researcher before maintainers see them.

GPT-6 Cyber by the Numbers

On ExploitGym, a benchmark that tests whether a model can exploit a specified vulnerability, GPT-6 Cyber scored 48.5% against 32.1% for GPT-5.6 Cyber. OpenAI calls that a 51% relative improvement. Source: OpenAI.

The Linux run in mid-October cost $1,213,430 in API credits, or $1,279 per working exploit. OpenAI says the newer model found more than five times as many new root exploits as a GPT-5.6 Sol run in July, despite three months of kernel hardening in between. Source: OpenAI.

81

New high-severity Firefox bugs (OpenAI)

948

Working Linux root exploits

48.5%

ExploitGym score, up from 32.1%

$1,279

API cost per Linux exploit

OpenAI also put a theoretical price on the haul. Using historical competition bounties of $20,000 to $70,000 per Linux root exploit, it estimates the 948 exploits at $18.9 million to $66.3 million. That is a hypothetical company calculation, not money anyone paid. Source: OpenAI.

MeasureGPT-5.6 Cyber / earlier runGPT-6 CyberSource
ExploitGym score32.1%48.5%OpenAI
Stayed in scope (no safety prompt)94.49%99.93%OpenAI
Linux root exploits, same harness166 (GPT-5.6 Sol)948OpenAI
Preparedness rating for cyberHighCriticalOpenAI
Company-reported comparisons; not independently verified.

Why OpenAI Calls the Model "Critical" Risk

OpenAI’s Preparedness Framework rates how dangerous a model’s abilities could be if misused. OpenAI classifies GPT-6 Cyber as Critical in cybersecurity, a step up from High for GPT-5.6 Cyber. Source: OpenAI. Source: BleepingComputer.

OpenAI says it paired the model with stronger controls. They include monitoring whether actions match a user’s authorization, required IP allowlisting for Daybreak Red API access, checks on stolen credentials, and new blocks on high-risk work against industrial control systems. Source: OpenAI.

The company also focused on keeping the model inside its assignment. In a test with deliberately easy-to-hack side services, GPT-6 Cyber avoided out-of-scope targets in 99.93% of runs without its deployment safety prompt, and in 100% with it. Source: OpenAI.

Critical infrastructure is a particular concern. OpenAI says the security firm Dragos tested Daybreak Red models on a mock power plant, where a model found two zero-day flaws in engineering software that handles protective relay data. Dragos chief AI officer Rob T. Lee warned that “the skill floor is dropping” for attack preparation. Source: OpenAI.

What It Means for Canada and the U.S.

For everyday users, the practical impact is patches. Canada’s Canadian Centre for Cyber Security flagged the October Mozilla fixes on October 28, listing Firefox versions before 159, Firefox ESR before 115.44 and 153.6, and Thunderbird before 159 and 153.6 as affected. Source: Canadian Centre for Cyber Security.

The Cyber Centre’s guidance is to review the advisories and apply any necessary updates as they become available. That applies equally to home users, businesses and public-sector offices on both sides of the border. Source: Canadian Centre for Cyber Security.

Linux matters well beyond desktop users. OpenAI notes the kernel underpins cloud infrastructure, Android devices and supercomputers. Consider a Canadian or U.S. business that rents a Linux cloud server and gives staff or contractors their own logins. Source: OpenAI.

OpenAI warns that, for now, a low-privileged account on an unpatched system should be treated as “functionally equivalent to root access” for a well-resourced attacker. In that example, any one compromised login could become full control of the server. Source: OpenAI.

Funding may also reach smaller defenders. OpenAI says its $1 billion Daybreak for Frontline Defenders grant program subsidizes access for resource-constrained teams such as government organizations and essential service providers. It has not published country-by-country eligibility in the sources reviewed here. Source: OpenAI.

This is part of a fast-moving pattern. Launch Len Media recently covered an OpenSSH password bypass on Alpine Linux and a U.S. warning about hacked grocery refrigeration controllers, both reminders that unpatched systems remain the easiest target.

A Fair Counterpoint

A fair counterpoint is that nearly every headline number here comes from OpenAI, which sells access to the model. The benchmarks, the 81 Firefox bugs and the Linux tally have not been independently audited in the sources reviewed for this article. Source: BleepingComputer.

The same tools can help attackers. BleepingComputer reports that Chinese hackers recently used an AI-powered penetration testing framework called ARTEX to automate attacks against more than 200 organizations, compromising at least nine. Source: BleepingComputer.

Finding bugs is also not the same as fixing them. OpenAI says it disclosed 304 critical Linux findings to maintainers with candidate patches and is still working on others, while volunteer maintainers must review the flood of reports. Source: OpenAI.

OpenAI itself notes that within 10 days of its scan, more than 100 Linux root bugs had been patched or staged by other contributors, with about 300 more under public discussion. That suggests the wider community is racing on the same problem. Source: OpenAI.

What Readers Can Do

The safest response to a surge in AI-found bugs is boring and effective: keep software current, and limit who can log in to your machines.

  • Update Firefox: open Help, then About Firefox, and confirm you are on version 159 or later; ESR users should be on 115.44 or 153.6 or later.
  • Update Thunderbird: the Cyber Centre listed Thunderbird before 159 and 153.6 in the same October advisory.
  • Patch Linux systems: apply your distribution’s kernel updates and reboot, especially on shared servers and cloud instances.
  • Limit local accounts: remove unused logins and avoid giving untrusted users shell access, since local root exploits need a foothold first.
  • Follow official alerts: the Canadian Centre for Cyber Security and Mozilla’s advisory pages publish fixes as they ship.

How We Reported This Story

This article is based on OpenAI’s GPT-6 Cyber announcement and its Linux research post, Mozilla’s Firefox 159 security advisory, the Canadian Centre for Cyber Security’s AV26-1100 advisory, and BleepingComputer’s report. Each figure links to the page it came from, and the piece went through an editorial fact-check against those pages before publication.

Performance and discovery figures are company-reported and labelled as such. Mozilla’s advisory credits 41 CVEs to OpenAI but does not name the model, so we do not claim those are the same as OpenAI’s 81 findings.

Frequently Asked Questions

What is GPT-6 Cyber?

GPT-6 Cyber is an OpenAI model built for security research, announced November 5, 2026. It finds and validates exploitable vulnerabilities and drafts patches, and is available only to approved Daybreak Red customers.

Is Firefox safe to use after these findings?

OpenAI says it reported the bugs to Mozilla with proposed patches, and Mozilla has shipped fixes for many of them. Keep Firefox updated to version 159 or later, or the matching ESR release, to get the published fixes.

Can anyone use GPT-6 Cyber?

No. OpenAI is offering it through its API and Codex to a select group of approved Daybreak Red customers, with plans to widen access to all Daybreak Red customers. IP allowlisting is required by default.

Why is OpenAI calling the model “Critical” risk?

Under OpenAI’s Preparedness Framework, Critical sits above the High rating given to GPT-5.6 Cyber. OpenAI says it based the rating on the model’s evaluation results and added stronger monitoring and access controls.

Should Linux users worry about the 948 exploits?

The exploits require an existing account on the machine, and OpenAI says it is disclosing findings with patches. Applying kernel updates promptly and limiting local accounts are the most useful steps.

Sources and Further Reading

This article was researched with AI assistance and checked against the sources listed above. Facts are current as of the publication date; follow the source links for updates.

Share This Article
Leave a Comment

Leave a Reply