Chrome 158 lists 204 fixes, 6 critical; 8 credits say assisted by Claude, 33 to 54 days before release
Chrome 158's notes of 17 November list 204 security fixes, 6 Critical. Eight credit lines say assisted by Claude and were reported 33 to 54 days before release, where Chrome 155's AI-named lines were 6 to 12 days and 14 in number.
By Parminder Kumar Sharma · LinkedIn · · 17 min read
Eight of the 204 credit lines say "assisted by Claude", reported 33 to 54 days before release
The Chrome Releases post of Tuesday 17 November 2026 promotes Chrome 158 to the stable channel and says the update "includes 204 security fixes". It lists each fix with a reward field, a bug number, a severity, a CVE id, a component and a credit line that ends in a report date. This site read the saved page on 18 November 2026.
Eight credit lines say "assisted by Claude". Their report dates fall 33, 35, 37, 37, 41, 52, 52 and 54 days before the 17 November release: a median of 39 days. For all 204 lines the median is 35 days, with a range of 8 to 661 (all derived: 17 November minus each printed report date, counted by script). Six weeks earlier, brief 261 counted the same kind of line in Chrome 155 and found fourteen AI-named lines, all reported 6 to 12 days before release, against a median of 97 days for all 247. In 158 the AI-named lines sit just above the middle of the list, not at its edge. Two of the six Critical lines carry the phrase (derived).
That is the checkable fact. Here is what it does not establish:
- A count of bugs found with AI. A credit line is the reporter's own statement as Google prints it. The page does not say what "assisted by" covered: finding, a proof of concept, triage or writing the report.
- That the other 4 Anthropic lines used no AI. 12 lines name Anthropic and 8 say "assisted by Claude". The other 4 name Anthropic without the phrase (derived). The page does not say whether they used a tool.
- That AI-named bugs are the severest. The only share the page supports is 7 of 68 Critical and High lines (10.3%, derived), against 14 of 57 (24.6%) in 155.
- When any fix landed in the source tree. The interval counted here runs from the report date to the stable release, not to the patch landing. The page does not say when each fix landed.
- That any of the 204 was exploited, or was not. The page prints no exploitation marker, no CVSS score and no bug detail: the notes say access to bug details "may be kept restricted until a majority of users are updated".
- How many users have updated. Not stated. The page says the release "will roll out over the coming days/weeks".
The same count six weeks later: fewer fixes, more Critical, fewer AI-named lines
Chrome 155 reached stable on 6 October 2026 and Chrome 158 on 17 November: 42 days later and three version numbers on (derived). This site read neither 156 nor 157, so what follows is a comparison of two releases, not a trend. Table 1 puts the two counts side by side, made with the same definitions except where the next paragraphs say otherwise.
Table 1. Chrome 155 (notes of 6 October 2026) against Chrome 158 (notes of 17 November 2026). Counts by script over each page; shares, changes and day counts are derived.
| Measure | Chrome 155 | Chrome 158 |
|---|---|---|
| Security fixes stated | 247 | 204 (43 fewer, 17.4%) |
| Critical, High, Medium, Low | 4, 53, 122, 68 | 6, 62, 82, 54 |
| Critical plus High | 57 (23.1%) | 68 (33.3%) |
| Credit lines naming an AI tool | 14 (5.7%) | 8 (3.9%) |
| Tools named | Claude 12, OpenAI Codex Security 2 | Claude 8, no other tool name |
| Credit lines naming Anthropic | 12 | 12 |
| AI-named lines among Critical and High | 14 of 57 (24.6%) | 7 of 68 (10.3%) |
| Report to release, AI-named lines | 6 to 12 days, median 8 | 33 to 54 days, median 39 |
| Report to release, all lines | median 97, range 5 to 680 | median 35, range 8 to 661 |
| Credited to Google | 185 (74.9%) | 145 (71.1%) |
| Reward field with a dollar amount | 16, totalling $33,000 | 5, totalling $11,000 |
Where the definition had to change. Three places, each small.
- Brief 261 counted one reporter's exact credit string plus the two OpenAI Codex Security lines. In 158 the phrase "assisted by Claude" appears under two different named reporters (7 lines and 1 line), so this brief counts credit lines that contain the phrase: 8. Applied to 155 the phrase gives the same 12 as the exact string did.
- Brief 261 also noted a wider reading that added a line crediting a team with "CodeBuddy Security" in its name. No such tool name appears in 158. Apart from "Claude", no tool name appears at all: this site searched the credit lines for OpenAI, Codex, Gemini, Big Sleep, Copilot, DeepMind, LLM, agent and the word "AI" and found none. The two OpenAI Codex Security lines of 155 have no counterpart in 158.
- Seven lines print only a name, the same name string as seven of the eight Claude lines, with no employer and no tool. As in 261, this brief does not count them as AI-named; whether they are the same person, or used a tool, is not stated. They are 1 Critical (use after free in GarbageCollection), 5 High and 1 Low, all use after free (derived).
Why "12 and 12" misleads. Twelve lines name Anthropic in both releases. In 155 all twelve said "assisted by Claude". In 158 eight do, one says "(Anthropic)" with no tool (the Critical use after free in WebAudio), and three credit "Anthropic and Ada Logics." on High lines, all "Incorrect authorization", two in Media and one in PerformanceAPIs, each with reward field [TBD]. Ada Logics is an organisation; the page does not say what either organisation did in those three reports or whether any tool was used.
The 155 speed pattern did not repeat
Brief 261's headline was speed: every AI-named line was reported within 12 days of release, while the median for all 247 was 97. Chrome 158 shows nothing like it. The eight Claude lines are 33 to 54 days out, the median of all 204 lines has fallen from 97 to 35, and the gap between the two medians has gone from 89 days shorter for the AI-named lines to 4 days longer (derived). The diagram puts every line on one axis.
Some figures behind the picture, all derived:
- Severity does not explain it. The 68 Critical and High lines have a median of 31 days; the 61 of them with no Claude credit have 29. The 7 Critical and High lines that say "assisted by Claude" are at 35 to 54 days. Brief 261 offered an inference that severe fixes are merged late into the stable branch and so ship soon after the report. In 155 that fitted: its 57 Critical and High lines had a median of 26 days. In 158 it fits less well, and the page says nothing either way.
- Google's own lines moved most. The 145 lines credited to "Google" have a median of 32 days and a range of 8 to 234, against 125 and 5 to 194 in 155.
- Short gaps are rarer. Three lines in 158 were reported 12 days or fewer before release (21 in 155). Eighteen were reported 180 days or more before (derived).
- Same weeks, different release. Brief 261's fourteen AI-named lines were all reported between 24 and 30 September and shipped on 6 October. Three of the eight Claude lines in 158 carry report dates of 24 and 26 September, inside that window, and shipped on 17 November: 52 to 54 days. Why a report from the same week went into one release and not the earlier one is not stated.
- Report dates range wide. The oldest Google line is dated 28 March 2026 (234 days). The oldest line of all is dated 25 January 2025 (661 days) and is a Low, credited to an outside name. Each interval runs from report to stable release, not to the patch landing in the repository.
None of this says anything about whether the fixes were quick or slow to write. It says only when the page records that Google was told, and when the release went out.
What each kind of credit line states
Table 2 groups the 204 credit lines by what they print. The grouping and counts are this site's; the credit strings are Google's. Researcher names are left out.
Table 2. The 204 credit lines of the Chrome 158 notes of 17 November 2026 by kind. Counts by script; severity counts are Google labels.
| Credit as printed (lines) | Stated on the page | Not stated |
|---|---|---|
| "Google" (145) | A company name only. 2 Critical, 36 High, 70 Medium, 37 Low. Reports dated 28 March to 9 November 2026 | Which team or tool, how many finders, any use of AI |
| A named reporter, "(Anthropic), assisted by Claude" (8) | A reporter, an employer and a named model that assisted. Two different reporters across the 8 lines. 2 Critical (GPU, Media), 5 High, 1 Low. 5 use after free, 2 buffer overflow, 1 out of bounds write | What the assistance was. How many of the 8 the model found |
| A named reporter, "(Anthropic)", no tool (1) | One Critical use after free in WebAudio | Whether a tool was used |
| "Anthropic and Ada Logics." (3) | Two organisations credited together on 3 High lines, all incorrect authorization, reward [TBD] | Which did what; whether any tool was used |
| A name only, the same string as seven of the eight Claude lines (7) | 1 Critical, 5 High, 1 Low, all use after free, reward [N/A]. Not counted as AI-named | Employer, tool, or that it is the same person |
| All other outside credits (40) | A name, handle or team; one line prints a name followed by "(Google)". 5 carry a dollar amount | Any method. None names an AI product, a model vendor or the word "AI" |
Three parties have an interest in how these lines are counted, and each is stated once. Google ships Chrome, sets the severity labels and writes the credit lines, and 145 of the 204 say only "Google". Anthropic sells Claude, which the page credits as an assisting tool on 8 lines, and Google's text names Anthropic on 12. OpenAI sells Codex Security, which the 155 notes credited twice and the 158 notes do not mention. None of that is evidence about any single report. Every figure here is Google's, counted by this site, and a credit line is a statement recorded by Google, not a finding by Chrome.
The AI-named lines are also a different mix from 155. Five of the eight Claude lines are use after free, against 12 of 12 in 155; the other three are two buffer overflows and an out of bounds write. Use after free is 51 of the 204 lines (25.0%) against 34 of 247 (13.8%) in 155 (derived). The page does not say whether the mix reflects what the tools find, what the reporter chose to target, or something else.
The reward field: [N/A], [TBD] and five dollar amounts
Every line opens with a bracket before the bug number. Table 3 counts what it holds. The page defines neither "[N/A]" nor "[TBD]", so what they mean is not stated.
Table 3. The reward field of the 204 lines in the Chrome 158 notes. Counts by script; the dollar total is derived.
| Reward field | Lines | Whose credit |
|---|---|---|
| [N/A] | 163 | 145 "Google", 8 "assisted by Claude", 7 name only, 1 Anthropic without a tool, 2 other outside credits |
| [TBD] | 36 | 3 "Anthropic and Ada Logics.", 33 other outside credits |
| A dollar amount | 5 ($4,000, $2,000 three times, $1,000: $11,000 in all) | All five outside credits: 1 High, 3 Medium, 1 Low. None to Google, Anthropic or the name-only string |
Chrome 155's notes printed 16 dollar amounts totalling $33,000, 41 [TBD] and 190 [N/A]. In 155 the Claude lines were 9 [TBD] and 3 [N/A]; in 158 all 8 are [N/A]. The page offers no reason for any of this.
This site's inference, not the page's: lines credited to Google's own teams, and to researchers employed by a lab, are likely outside a public reward programme, which would fit [N/A] on all 145 Google lines and on every line crediting Anthropic except the three that read [TBD]. The page does not say so, and "likely" is the strongest word the evidence supports.
The six Critical lines and the mix behind them
Table 4 lists the six Critical lines exactly as the page orders them, with the credit as printed and the days from report to release (derived). The CVE ids are public and printed on the page; the page prints no bug detail, no indicator and no hash, and this site adds none. Two of the six (one third, derived) credit a report that says "assisted by Claude". Three of the six name Anthropic, and four print the same name string as the Anthropic lines. Two are credited to Google. All six show [N/A] in the reward field.
Table 4. The six Critical lines of the Chrome 158 notes of 17 November 2026. CVE, type and component as printed; credit grouped by this site with names left out; days are 17 November 2026 minus the report date (derived).
| CVE and weakness as printed | Credit | Reported | Days |
|---|---|---|---|
| CVE-2026-130963: Use after free in WebView | Reported by Google | 15 September 2026 | 63 |
| CVE-2026-131031: Out of bounds write in GPU | A named reporter, Anthropic, assisted by Claude | 7 October 2026 | 41 |
| CVE-2026-131060: Buffer overflow in Media | A named reporter, Anthropic, assisted by Claude | 13 October 2026 | 35 |
| CVE-2026-131008: Use after free in GarbageCollection | A name only (the same string as the Anthropic lines) | 17 October 2026 | 31 |
| CVE-2026-131054: Out of bounds write in ANGLE | Reported by Google | 22 October 2026 | 26 |
| CVE-2026-130975: Use after free in WebAudio | A named reporter, Anthropic, no tool named | 26 October 2026 | 22 |
The mix behind them, counted over the text before " in " in each entry (derived; 28 distinct weakness labels and 92 distinct components):
- Weakness labels, top ten: use after free 51, incorrect authorization 34, missing authorization 16, uninitialized resource 14, information leak 11, ui misrepresentation 10, out of bounds read 9, improper input validation 9, race condition 8, confused deputy 7. The two authorization labels together are 50 lines (24.5%).
- Components, top ten: DevTools 12, ANGLE 10, Autofill 10, Media 9, GPU 8, WebAppInstalls 8, Extensions 7, V8 7, Editing 6, Mobile 5. Component names are Google's labels and mix browser parts, graphics code and platform features.
- Severity by credit group. Google 2 Critical, 36 High, 70 Medium and 37 Low. The 8 Claude lines are 2 Critical, 5 High and 1 Low; none is Medium.
The page prints no indicators or hashes, no exploitation marker and no CVSS score. A reader who wants a score has to look each CVE up at the CVE Program or the National Vulnerability Database, which this site did not do, and should attribute any score to whoever assigned it.
What the Chrome Releases notes state, and what they leave out
Table 5. Stated and not stated in the Chrome 158 desktop notes and the Android entry on the blog front page, as read on 18 November 2026.
| Topic | Stated | Not stated |
|---|---|---|
| Fix count and severity | "204 security fixes"; 6 Critical, 62 High, 82 Medium, 54 Low by script. Severity is Google's label | A CVSS score for any line; the rule behind each label |
| Versions and rollout | 158.0.8127.36 for Linux, 158.0.8127.36 and .37 for Windows and Mac; rolling out "over the coming days/weeks" | How many users have the update; when a given machine receives it |
| Android | Chrome 158 (158.0.8127.42) for Android, on Google Play "over the next few days", with "the same security fixes" as the desktop release "unless otherwise noted" | Which, if any, desktop fixes the Android release does not carry |
| Credit | One line per fix: reporter text and a report date. 8 say "assisted by Claude"; 12 name Anthropic | What any tool did; how many of the 204 involved AI; how credit text is chosen |
| Reward | [N/A], [TBD] or a dollar amount on each line | What [N/A] and [TBD] mean |
| Bug detail | A note that details and links may stay restricted until most users are updated | Any bug detail for any of the 204 |
| Exploitation | No sentence about it, on any line | Whether any of the 204 was exploited |
| Timing | A report date per line | When each fix landed in the source tree; the stable branch date |
Differences inside and between the records, unresolved
The records leave five differences, which this brief reports and does not resolve.
- A lab and a tool. 12 lines name Anthropic and 8 name Claude. The page does not say whether the other 4 used AI.
- The reward column. 5 lines show a dollar amount and the rest show [N/A] or [TBD]. The page does not define the markers. The explanation in the previous section is this site's inference.
- Restricted detail. The release says bug details and links may stay restricted until most users are updated, so no bug detail is checkable from the page.
- Desktop and Android. The desktop post covers 158.0.8127.36 and .37. The Android entry on the front page says Android releases carry the same security fixes "unless otherwise noted"; this site read no separate Android security list.
- Report dates and landing. Report dates run from 25 January 2025 to 9 November 2026. A fix can be months old when it ships, and the interval counted here is from report to stable release, not to the patch landing in the repository. The page does not say when each fix landed.
What to do, in order
The page says the release "will roll out over the coming days/weeks", so some machines will not have it yet. The items below are this site's suggestions, not Google's advice.
Take this with you
This site's suggestions
- Update Chrome, and every Chromium based browser and embedded runtime (Electron, WebView, Edge, Brave and others), to a version based on Chrome 158. Check each vendor's own release notes, which this site did not read.
- Enterprise: use the managed update policy and check the browser version across the fleet, including Android devices on 158.0.8127.42.
- Chrome applies an update at relaunch: a machine that has not restarted may still run the old version, so record the running version, not only the installed one.
- Watch the vendors' own security bulletins for the six Critical CVE ids in Table 4 and for the other 198 ids on the page.
- For security teams measuring AI in vulnerability reports: record, in your own bug bounty or intake, the share of reports that state AI assistance, the days from report to fix and the severity, as the Chrome page lets anyone do for its own list.
- Read the headline number for what it is: 204 fixes is a count of reported bugs fixed in one release, not a count of attacks.
The question the credit line leaves open
For two releases Google's notes have carried the only public record of how some of these bugs were found: a few words after "Reported by". In 155 those words named a tool on 14 lines and the lines were fast. In 158 they name a tool on 8 and the lines are ordinary. Meanwhile 145 lines say "Google", and the credited name on seven more lines carries no employer. When the credit line is the only record of how a bug was found, who counts how many were found by machines?
Key facts
Sources
- PrimaryStable Channel Update for Desktop, Tuesday 17 November 2026: Chrome 158.0.8127.36 for Linux, 158.0.8127.36 and .37 for Windows and Mac, 204 security fixes, one line per fix with reward field, bug number, severity, CVE id, component and credit line with report date. Read in full by this site on 18 November 2026; every count in the brief is a script count over itGoogle Chrome Releasesaccessed 2026-11-18
- PrimaryThe blog's front page as read on 18 November 2026: carries the Chrome for Android update of 17 November (158.0.8127.42, same security fixes as the desktop release unless otherwise noted) above the desktop postGoogle Chrome Releasesaccessed 2026-11-18


