Josip Šokčević

Chrome's 2026 CVE Spike: Is Agentic the Reason?

Chrome runs on roughly 65% of the world’s browsers. CVE counts for 2021–2025 were on a steady downward trend — then 2026 broke the pattern sharply. I scraped five years of data from the Chrome Releases blog and the Chromium Dash API to see what’s behind it.

This post was generated by AI. The analysis was done for business purposes and the data is shared here as it may be useful to others. Data covers Stable Linux releases, Jan 2021–Jun 2026. CVE severity is extracted from blog post HTML and may undercount — treat as directional. CVE counts include all platforms, not just Linux.


Background: Release Cadence

Chrome switched from a 6-week to a 4-week major release cycle with M94 in September 2021. Average days between security-focused updates: 11.2 (2021) → 4.1 (2026).

Year Stable Releases Security Releases Avg Days Between Milestones
2021 33 31 (8.1%) 11.2 87–96
2022 37 63 (12.0%) 6.4 97–108
2023 44 71 (13.0%) 5.6 109–120
2024 53 83 (14.3%) 5.1 120–131
2025 50 81 (14.6%) 5.1 131–143
2026 28 48 (16.5%) 4.1 143–149 (Jan–Jun)

The CVE Trend Break

CVE counts per security release were declining steadily from 2021 to 2025. Then 2026:

Year Total CVEs Avg CVEs / Security Release
2021 353 11.4
2022 562 8.9
2023 450 6.3
2024 420 5.1
2025 381 4.7
2026 1,589 33.1 (6 months)

The monthly breakdown shows this isn’t a single outlier patch — it’s an acceleration across the quarter:

Month Sec Releases CVEs Critical High Medium Low
Jan 7 41 0 10 5 2
Feb 8 52 0 9 3 0
Mar 12 130 0 8 0 0
Apr 8 232 12 77 30 24
May 7 547 48 273 181 42
Jun* 6 587 63 199 230 94

June through June 13 (~half the month).

Jan–Mar averaged 74 CVEs/month. Apr–Jun is averaging ~455 CVEs/month — a 6× step-up within one quarter.


What Changed

The vulnerability type mix shifted alongside the volume increase. The all-time breakdown (2021–2026):

Vulnerability Type Count Share
Use after free 1,220 32%
Other / unclassified 715 19%
Inappropriate implementation 517 14%
Out of bounds read/write 278 7%
Heap buffer overflow 226 6%
Insufficient validation 208 6%
Insufficient policy enforcement 189 5%
Type confusion 171 5%
Integer overflow 88 2%
Incorrect security UI 61 2%

In early 2026 (Jan–Mar), the dominant types were “Inappropriate implementation” and unclassified. From April onward, “Insufficient validation” surged — 110 CVEs in the first half of June alone. UAF, which has always led the charts, roughly tripled in volume from its 2025 rate.

The June 12 patch (M149) is the largest single update in this dataset: 17 Critical CVEs across Ozone, Views, TabStrip, FileSystem, Printing, Chromoting, GPU, and libyuv. Ozone alone accounts for at least five distinct Critical CVEs since April.

Notably, the number of security releases hasn’t increased proportionally — from 81 in all of 2025 to ~96 projected for 2026. It’s CVEs per release that jumped, from 4.7 to 33.1. More bugs are being found per audit cycle, not just more audit cycles.

One open question: is improved LLM and agentic tooling in security research responsible for the volume increase? AI-assisted fuzzing and automated code analysis have advanced considerably in the past year, and UAF bugs in C++ are exactly the class these tools are well-suited to surface at scale. The data doesn’t answer this — but the timing and the type distribution are consistent with the hypothesis.


EoY 2026 Projection

Scenario Basis Est. Full-Year CVEs Est. Security Releases
Conservative Jan–May rate (~200/mo) ~2,400 ~100
Trend-based Apr–Jun rate (~455/mo) ~3,900 ~100

Stable Linux releases are on track for ~54–56 for the full year (2025: 50).


Sources: Chromium Dash API, Chrome Releases Blog RSS.