Firefox 150 Ships With 271 AI-Found Vulnerabilities Patched, as Mozilla Declares Defenders Can Finally Win
Mozilla released Firefox 150 on April 21, 2026, fixing 271 vulnerabilities surfaced by Anthropic's Claude Mythos Preview in a security sweep Mozilla's CTO calls a turning point for defender-side AI.
Overview
Mozilla released Firefox 150 on April 21, 2026, shipping a wave of new user features alongside an unusually large security payload: 271 vulnerabilities identified and fixed with the help of Anthropic’s Claude Mythos Preview, an agentic model applied to the browser’s codebase as part of a collaboration tied to Anthropic’s Project Glasswing cybersecurity effort. According to Engadget, Mozilla used the Mythos Preview to surface bugs across Firefox and then patched them before shipping 150.
Mozilla CTO Bobby Holley framed the results as a shift in the long-standing asymmetry between attackers and defenders in a companion Mozilla blog post titled “The zero-days are numbered” published alongside the release.
This follows earlier Machine Herald coverage of Anthropic’s cyber guardrails testing ahead of Mythos’s broader rollout.
What We Know
Firefox 150 is a stable release shipped on April 21, 2026. Engadget reports that Mozilla said it patched 271 Firefox vulnerabilities identified by Anthropic’s Claude Mythos, quoting the foundation’s statement that “so far we’ve found no category or complexity of vulnerability that humans can find that this model can’t.”
In its own blog post, Mozilla describes the work as applying “an early version of Claude Mythos Preview to Firefox” and confirms that “this week’s release of Firefox 150 includes fixes for 271 vulnerabilities identified during this initial evaluation.” Engadget characterizes Project Glasswing as Anthropic’s cybersecurity initiative, announced earlier in April 2026, that “was met with plenty of skepticism” — and frames Mozilla’s third-party results as the first significant external validation of the approach.
The Register reports that Mozilla CTO Bobby Holley acknowledged the findings initially gave the team “vertigo” but framed them as a decisive moment for defenders. Holley wrote that “defenders finally have a chance to win, decisively,” arguing that industry security work had previously “fought security to a draw” by making exploits expensive enough to deter most threat actors. The same article notes Holley’s claim that Mythos matches elite human researchers’ capabilities rather than exceeding them, writing that Mozilla “haven’t seen any bugs that couldn’t have been found by an elite human researcher.”
Slashdot carried the announcement, summarizing Mozilla’s claim that Mythos produced 271 actionable bug reports and noting Holley’s framing that the work tips the long-running balance between attackers and defenders.
User-Facing Framing
Despite the scale of the security payload, Mozilla’s public framing kept an emphasis on user agency. Engadget observed that Mozilla preserved user choice by allowing users to disable AI features, continuing the opt-out posture the browser has taken throughout its AI integration cycle. The same coverage treats Firefox’s endorsement of Mythos as notable precisely because it is a third-party validation rather than an Anthropic self-report.
What We Don’t Know
Mozilla has not published a public per-CVE breakdown tying each of the 271 fixes to Mythos’s output. The Register’s coverage does not enumerate the individual bugs, and Engadget notes only that the 271 count reflects Mozilla’s own tally from the initial Mythos evaluation. How the number breaks down across critical, high, moderate, and low severity tiers — or across components like the JavaScript engine, WebAssembly runtime, DOM, and graphics stack — has not been disclosed publicly.
It is also unclear how Project Glasswing will scale. Anthropic has not publicly disclosed which other organizations currently have Claude Mythos Preview access, under what terms, or what eligibility criteria govern Project Glasswing participation; none of the cited reports (Engadget, The Register, Mozilla’s own blog post) specify cost structures or whether smaller open-source projects will receive comparable access. The degree to which Mythos’s findings reflect Firefox-specific codebase characteristics — rather than generalizable AI bug-hunting capability — remains to be seen as more vendors publish their own audit results.
Finally, Mozilla’s optimistic framing depends on defenders keeping a material lead in model access. The Register notes Holley’s own caveat that Mythos has not surfaced bug categories beyond what elite human researchers could find, leaving open the possibility that adversaries with comparable tooling could simply match, not be outpaced by, defensive teams.
Why It Matters
For users, Firefox 150 is a routine update to install. For the software industry, it is the most concrete public example to date of an AI model being used to meaningfully compress a mature browser’s outstanding vulnerability backlog. Mozilla’s decision to publish a count rather than just individual CVEs — and to frame the result in terms of the defender/attacker asymmetry — positions the release as a reference point for how AI-assisted security audits are reported going forward.
Whether Holley’s “zero-days are numbered” thesis survives contact with adversarial AI use of equivalent models is the question Project Glasswing’s next round of disclosures will have to answer.