What A Downloadable AI Did With Firefox Fixes—and Why It Matters
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: What A Downloadable AI Did With Firefox Fixes—and Why It Matters on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the little things that make your day delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Epoch AI researchers report that open-weight model GLM 5.3 built working attacks against six Firefox releases by analyzing public security fixes, with the fastest run taking 12 hours. The lab attacks also depended on an unpublished flaw, which Epoch reported and Mozilla patched in Firefox 159. The study did not test attacks on real users, and Epoch has not yet tested whether the Firefox 159 fix stops similar attacks.

Open-weight AI model GLM 5.3 built working attacks against six recent Firefox releases by studying public security fixes, according to a laboratory report published October 29 by Epoch AI. The fastest attempt took 12 hours; the attacks also used an unpublished vulnerability that Epoch reported to Mozilla, which patched it in Firefox 159, released October 27.

Epoch’s six researchers tested Firefox releases 156.0 through 158.0.1, issued from September 17 to October 26. They gave the model each version number and public information available at that release, then tasked it with creating a webpage that could read a local file when opened in the specified browser. The researchers allowed up to 48 hours for each attempt; the fastest successful run took 12 hours. They estimate the API cost at about $1,000 per version once the test setup was in place.

The model identified security fixes in public code before those fixes reached users of the stable release. Epoch says GLM 5.3 selected bugs, built a multi-stage attack and worked through failures without researchers guiding its choices. In each successful test against recent versions, the attack also used a separate, unpublished flaw to get past Firefox’s sandbox, a feature intended to limit what a compromised webpage can access. Epoch disclosed that flaw to Mozilla.

Mozilla released Firefox 159 on October 27, the day before Epoch’s report appeared. The source material says the release addressed the reported flaw. It also cites Mozilla’s security advisory, which lists 167 fixes: 108 rated high, 52 moderate and seven low. The test’s requested outcome was reading a file, not taking control of the entire computer. Epoch says it spent more than $100,000 building and testing its setup, and its attempt to apply the same setup to Chrome failed after about that amount of effort.

At a glance
reportWhen: Epoch AI report published October 29, 2…
The developmentEpoch AI published a report saying a downloadable AI model turned public Firefox security fixes into working attacks against six releases, while Mozilla patched an additional flaw the attacks used.
Thorsten Meyer AI
Lab test · Epoch AI

Infographic 01 · The patch gap

A published fix is a map. GLM 5.3 read it.

How an open-weight model turned Firefox security fixes into a working attack before the update reached users, in Epoch AI’s offline lab test.

6 / 6Firefox releases compromised in Epoch’s test (156.0 to 158.0.1)
12 hfastest successful run; up to 48 h were allowed
≈ $1,000marginal API cost per Firefox version, once the setup existed
0refusals or safeguards triggered, at any provider

Where the attack happens

The life of a security fix, simplified

  1. 1 · Fix writtenDevelopers find and fix a security bug
  2. 2 · Fix visibleThe change shows up in public code and test builds, disguised as an ordinary change. The patch gap opens
  3. 3 · Update shipsThe fixed version is released to everyone
  4. 4 · You install and restartOnly now is your browser protected. The gap closes for you

GLM 5.3 worked in step 2: it picked out the security fixes and turned them into a working attack before step 3.

Compromised in Epoch’s test

Released 17 Sep to 26 Oct 2026 · only information public at release

Firefox 156.0Firefox 156.0.1Firefox 157.0Firefox 157.0.1Firefox 158.0Firefox 158.0.1

Result: a web page that read a local file and sent it to the attacker’s server when visited. No further user action.

Firefox 159.0 · 27 Oct

Fixes the unpublished flaw every recent success relied on

Key flaw fixedNot yet re-tested by Epoch

Epoch expects the fix to make similar attacks “at least substantially more difficult”.

Sources: Epoch AI, 29 Oct 2026; Firefox 159.0 release notes. Lab test without internet access; no real users were attacked. No exploit details are shown here.
Powered by Thorsten Meyer AI

Public Fixes Can Aid Attackers

The report highlights the time between a security fix becoming public and users installing the update. In open-source software, code changes can reveal details about a repaired weakness. That information helps defenders understand what changed, but it can also give attackers clues about where to look. The window between disclosure and widespread updating is often called the patch gap.

Epoch’s finding is that an open-weight model could carry out much of the work in this particular lab task, not that every public fix can be turned into an attack or that real-world exploitation has been established. Still, a process that took as little as 12 hours in the test puts pressure on the time available for users and organizations to apply browser updates. The report also shows why prompt patching matters even when a vulnerability’s repair is already public: systems that have not received the update may remain exposed.

For users, the practical step is to update Firefox to version 159 or later and restart it. For organizations, the report points to patch deployment and restart policies as part of security response, alongside monitoring for newly disclosed flaws. Epoch’s test does not establish that any user was targeted or that the attack chain was used outside its controlled environment.

Products Worth Considering

Philips Sonicare DiamondClean 9500 Electric Toothbrush, Black, with 3 Brush Heads

Philips Sonicare DiamondClean 9500 Electric Toothbrush, Black, with 3 Brush Heads

OMRON Complete 2-in-1 Upper Arm Blood Pressure Monitor with EKG

OMRON Complete 2-in-1 Upper Arm Blood Pressure Monitor with EKG

As an affiliate, we earn on qualifying purchases.

How the Patch Gap Works

Software fixes can become visible before all users receive them. Firefox is an open-source project, so changes to its public code may offer clues about a corrected vulnerability. Test builds can also expose fixes before a stable update is installed across a user base. During that interval, attackers may compare versions or inspect code to infer what was repaired.

Epoch’s experiment focused on whether a model could use that public information with limited direct guidance. The researchers provided a test machine running Ubuntu 26.04 with default settings and an official Firefox release. They did not give the model internet access, and the setup could not contact real websites or people. The model was told to examine fixes not yet present in the stable release, then was allowed to work without further researcher input.

The report describes a repeated result across six releases, but not a broad survey of browsers or software. Epoch says earlier Firefox versions are likely also vulnerable, but it did not systematically test them. It also says the same approach failed against Chrome after roughly $100,000 in attempts. That result is specific to the researchers’ setup and does not establish that other browsers are safe from similar methods.

“widely available open-weight models can now do much, or even all, of this work on their own.”

— Epoch AI researchers

Firefox 159 Still Needs Testing

Epoch has not reported testing whether the Firefox 159 patch prevents the attack chain or how much it changes the effort required. Its expectation that attacks will become “at least substantially more difficult” is a forecast, not a demonstrated result. The report also does not show that the lab attack was used against real Firefox users.

The test covered six releases and one specified file-reading objective. It does not establish how often attackers could reproduce the work, whether other systems or browser configurations are vulnerable, or whether the model’s performance would carry over to different targets. Epoch says older Firefox releases are likely vulnerable, but it did not test them systematically. The supplied material also does not establish whether other AI providers or models would behave the same way.

Update Firefox and Check Results

Firefox users should install version 159 or later and restart the browser, according to the source report. Organizations will need to confirm that updates have reached managed devices, since downloading an update without restarting may leave the older browser session in use.

The next technical milestone is a test of Firefox 159 against the reported method. Epoch had not completed that assessment when its report was published. Further information from Mozilla or Epoch could clarify whether the patch blocks the chain, how the fix affects similar techniques and whether additional findings emerge. Until then, the confirmed result is limited to Epoch’s controlled tests on earlier releases and Mozilla’s release of a patch for the reported unpublished flaw.

Key Questions

What did GLM 5.3 do in the Firefox test?

Epoch AI says the model built webpages that could read a local file when opened in each of six tested Firefox releases. The work took up to 48 hours per attempt, with the fastest successful run taking 12 hours.

Did the test show that Firefox users were attacked?

No. Epoch described a controlled lab test with no internet access, so the setup could not attack real websites or people. The report does not establish that anyone used the method against real users.

Why did public security fixes help the model?

The report says GLM 5.3 examined public code changes that disclosed clues about bugs fixed in later Firefox releases. Those clues gave the model information it could use while targeting versions that did not yet include the fixes.

Did Mozilla fix the flaw used in the attacks?

Epoch reported the unpublished flaw to Mozilla, and the source material says Mozilla patched it in Firefox 159, released October 27. Epoch had not yet tested whether that release stops similar attacks.

What should Firefox users do?

Install Firefox 159 or later and restart the browser. The report does not say that users who update are protected from every possible browser vulnerability; it addresses the flaws and attack chain described in Epoch’s test.

Source: ThorstenMeyerAI.com

You May Also Like

AI Agent Access: Identity, Permissions, And Security

IdeaNavigator AI outlines a proposed access-control broker for AI agents, with short-lived credentials, task-level policies and revocation.

2026’S AI Revolution: 7 Developments To Watch

Explore the top 7 confirmed AI advancements expected in 2026, their significance, and what remains uncertain as the AI landscape rapidly evolves.

RHEO On The Web: Find Your Flow

Discover RHEO’s web version—an instant, private, browser-based fluid playground designed for calm, creativity, and mindfulness without downloads or sign-up.

The Internet’s AI Security Record, Examined Through Recent Headlines

Recent disclosures involving PHP, weak certificate keys, cloud data and software packages highlight separate security failures and urgent patching needs.