TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Epoch AI reports that open-weight model GLM 5.3 built file-reading attacks against six Firefox releases by analyzing public security fixes before users received updates. Mozilla patched an additional flaw used in the tests in Firefox 159, released October 27; the experiment was conducted in a lab and did not target real users.
Epoch AI reported October 29 that open-weight model GLM 5.3 built working attacks against six Firefox releases by analyzing publicly visible security fixes before those fixes reached users. Mozilla released Firefox 159 on October 27 with a patch for an additional flaw the researchers reported; Epoch’s tests took place in an offline lab, not against real users.
Epoch’s researchers tested Firefox versions 156.0 through 158.0.1, released between September 17 and October 26, according to the report. They gave the model a specific task: create a webpage that could read a local file when opened in the targeted Firefox version. The test environment used Ubuntu 26.04, default settings and Firefox’s official releases. Epoch said the model had no internet access, limiting the possibility of contacting real websites or people.
The researchers allowed up to 48 hours per attempt; the fastest successful run took 12 hours. They estimate that once the research setup was built, each additional Firefox version cost about $1,000 in API fees. Epoch says development and testing of the setup cost more than $100,000. Those figures describe this experiment and should not be read as a general cost estimate for attacks.
According to Epoch, the model identified fixes already visible in public code but not yet included in stable releases, then combined that information with a second, unpublished flaw. The researchers reported that additional flaw to Mozilla, which patched it in Firefox 159.0. Epoch also says the model did not refuse the task or encounter safeguards from any provider during the experiment. That is the researchers’ account of this specific test, not a broad finding about every model or deployment.
Security report · October 29, 2026
5 Takeaways From Epoch AI’s Test of Downloadable AI and Firefox Fixes
Epoch AI says open-weight model GLM 5.3 built file-reading attacks against six Firefox releases by studying public fixes before users received updates. The experiment ran offline in a lab and did not target real users.
The reported capability
“Indistinguishable from those of a real attacker”— Epoch AI researchers
Five key points
What the test showed
These findings describe one model, target, and controlled setup. They are not general estimates of attack cost or proof of real-world exploitation.
Patch details can expose clues
GLM 5.3 analyzed fixes visible in public code before they appeared in stable Firefox releases.
The task was file reading
The requested outcome was a webpage that could read a local file in a targeted Firefox version—not computer takeover.
Researchers reported another bug
Epoch says the model combined public fix information with an unpublished flaw, which Mozilla patched in Firefox 159.0.
The test was offline
The Ubuntu environment had no internet access. Epoch says the experiment did not involve real users.
Update and restart Firefox
Install Firefox 159 or later and restart the browser so the security update takes effect.
One report, specific setup
The findings do not show how reliably other models or environments could reproduce the results.
Why patch timing matters
The Firefox patch gap
A patch gap is the interval between a security fix becoming publicly visible and users installing the corresponding update. Public code can reveal where a bug was, while devices still run older versions. Epoch’s test asked whether an AI model could use that information to help develop an attack.
Tested releases ran from September 17 to October 26, 2026.
Researchers allowed this much time per attempt; the fastest success took 12 hours.
Ubuntu 26.04, default settings, and official Firefox releases, according to Epoch.
Reading the evidence
What the report does—and doesn’t—establish
Epoch reported a controlled experiment. Its results should be interpreted within the tested conditions.
Reported in the test
- GLM 5.3 built file-reading attacks against six Firefox releases.
- The model selected fixes, tested approaches, and adjusted after crashes, per Epoch.
- Mozilla included a patch for the additional flaw in Firefox 159.
Not established by the test
- That this method was used against Firefox users in the wild.
- How many users may have been exposed or whether files left the lab.
- That the result generalizes to other models, software, or environments.
Update to 159 or later
Then restart the browser so the update takes effect.
Organizations can check that managed devices received the release and review how quickly their update process closes the time between a public fix and installation.
Epoch also reported that its same setup failed against Chrome after about $100,000 in attempts. That result applies to this experiment and does not establish that Chrome is generally safe from AI-assisted vulnerability research.
Quick answers
Key questions
Which versions were tested?
Firefox 156.0, 156.0.1, 157.0, 157.0.1, 158.0, and 158.0.1, released from September 17 to October 26, 2026.
Did the test attack real users?
No. Epoch described an offline lab with no internet access. The report does not establish real-world use.
Did GLM 5.3 take over a computer?
The demonstrated capability was reading a local file through a webpage. The supplied report does not claim computer takeover.
What should users do?
Install Firefox 159 or later and restart the browser. Mozilla included the reported additional flaw’s patch in Firefox 159.0.
When Public Fixes Guide Attackers
The report highlights how the period between publishing a software fix and users installing it can expose systems. A public patch may reveal where a bug was, giving attackers information they can study while devices remain on older versions. Epoch’s result suggests that an open-weight model, in its test environment, could automate substantial parts of that analysis and attack development.
The demonstrated impact was reading a local file through a webpage, not taking control of a computer. Epoch said such access could expose sensitive material, including saved passwords, login cookies or private documents, depending on what was stored and accessible. The report does not establish that anyone used this method against Firefox users in the wild. For users and organizations, the immediate practical step is to install Firefox 159 or later and restart the browser so the update takes effect.
Products Worth Considering
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
The Firefox Patch Gap
The patch gap is the interval between a security fix becoming publicly visible and users receiving or installing the corresponding software update. In open-source projects, fixes can appear in public code; test builds may also reveal changes before a stable release. The gap predates this experiment. Epoch’s report focuses on whether an AI model could use that public information quickly and with limited human guidance.
Epoch said it supplied GLM 5.3 with the Firefox version, basic tools and public information available when each release appeared. It instructed the model to examine fixes not yet present in the stable release, then let it work without further input. The researchers reported that it selected fixes, tested approaches and adjusted after crashes. They also said they tested Chrome, but the same setup failed after about $100,000 in attempts. That result is specific to their setup and does not establish that Chrome is generally safe from AI-assisted vulnerability research.
“widely available open-weight models can now do much, or even all, of this work on their own.”
— Epoch AI researchers
Limits of the Lab Results
The report describes a controlled lab experiment, not evidence of attacks on people or confirmation that the method has been used in the wild. It tested six releases systematically; Epoch said earlier versions were “likely also vulnerable”, but did not test them systematically. The report also does not show how reliably other models, teams or real-world environments could reproduce the results.
Epoch’s findings depend on its chosen target, machine configuration, instructions and tools. The supplied account does not establish how many users might have been exposed before Firefox 159, whether any files were actually taken outside the lab, or how the model’s performance would change with different safeguards. The researchers’ report says the attack could read files; it does not claim that the model took over the computer.
Updating Firefox and Monitoring Risks
Firefox users should update to version 159 or later and restart the browser. Organizations can check whether managed devices have received the release and consider how quickly their update process closes the period between a fix becoming public and its installation.
Epoch’s report adds evidence to ongoing debate about how open-weight AI affects security research and misuse risks. Further work would need to test other software, models and operating conditions, and establish whether similar methods are being used outside controlled research. The supplied source does not identify a date for such follow-up or report a known real-world incident.
Key Questions
What did Epoch AI report?
Epoch AI said GLM 5.3 built file-reading attacks against six Firefox releases in an offline lab test by analyzing public security fixes and using an additional unpublished flaw.
Which Firefox versions were tested?
The report covers versions 156.0, 156.0.1, 157.0, 157.0.1, 158.0 and 158.0.1, released from September 17 to October 26, 2026.
Did the test attack real Firefox users?
No. Epoch described an offline lab environment with no internet access. The report does not establish that the method was used against real users.
What should Firefox users do?
Install Firefox 159 or later and restart the browser. Mozilla included a patch for the additional flaw reported by Epoch in Firefox 159.0.
Did GLM 5.3 take over a computer?
The demonstrated capability was reading a local file through a webpage. The supplied report says the attack did not take over the computer.
Source: Thorsten Meyer AI
