Desktop Chrome 157 fixes 213 flaws, seven critical, and a restart applies the update
Google moved Chrome 157 to the stable channel for Windows, macOS and Linux on 3 November, with 213 security fixes, seven of them rated critical. The fixed builds are 157.0.8103.38/.39 for Windows and macOS and 157.0.8103.38 for Linux. Google says the release will roll out over the following days and weeks.
PCWorld’s headline figure of over 200 is a rounded form of that 213 total, not a separate count. The more useful distinction is between a large security release and a confirmed attack. SecurityWeek reported that Google made no mention of any flaw being exploited in the wild, and reported no announcement of a vulnerability confirmed as exploited. That absence is not proof that exploitation never occurred, but it changes the immediate question from incident response to patch application.
Where the seven critical flaws sit
The seven critical flaws sit across graphics, rendering and core browser code. Google lists two buffer overflows in Dawn (CVE-2026-121156 and CVE-2026-121247), memory corruption in ANGLE (CVE-2026-121229), and use-after-free bugs in WebGL (CVE-2026-121246), Prerender (CVE-2026-121035), Geometry (CVE-2026-121203) and Core (CVE-2026-121126).
The public record is most detailed for CVE-2026-121156. It describes a heap-based buffer overflow in Chrome before 157.0.8103.38 that could potentially allow a remote attacker to execute arbitrary code outside the sandbox through a crafted HTML page. NVD records Chromium’s severity as critical and a CVSS 3.1 base score of 9.6 contributed by CISA-ADP, while NVD’s own assessment is not yet provided. The 9.6 is a contributor’s score rather than NVD’s own assessment, and no comparable score was available for the other six. The public descriptions do not establish a shared root cause or equivalent impact across all seven.
Critical is a severity label, not a measure of exploitation. The seven ratings indicate potential impact as assessed by the vendor; they do not show that attackers hold working exploits or that every flaw is equally dangerous. That is why the release can be both routine in form and worth prioritising in practice. The case for prompt action rests on the number of fixes and the severity ratings, not on an observed campaign.
Counts, categories and exposure
Beyond the critical group, SecurityWeek reported 77 high, 75 medium and 54 low severity fixes, which together with the seven critical issues account for the full 213. SecurityWeek lists incorrect or missing authorisation at 51, use-after-free at 45, UI misrepresentation at 19, type confusion and information leak at 13 each, out-of-bounds issues at 11, and improper input validation and uninitialised resource at nine each. A PCWorld account syndicated by Apple News summarised the most common types differently, putting use-after-free at 45 and data leaks and type mix-ups at 13 each.
Exposure is specific to stable-channel desktop Chrome before the fixed builds. The Canadian Centre for Cyber Security lists versions prior to 157.0.8103.38/.39 on Windows and macOS and prior to 157.0.8103.38 on Linux as affected. GovCERT.HK lists Chrome prior to 157.0.8103.38 and describes potential impacts including remote code execution, denial of service, elevation of privilege, information disclosure, security restriction bypass, spoofing and tampering. Those are potential consequences of successful exploitation, not a record of an attack.
What users and administrators should do
Chrome usually updates automatically, but downloading an update and running the fixed build are different states. GovCERT.HK advised users to relaunch Chrome to make the update effective. Reporting shows that a restart is required to fully apply the update and that users can check manually through chrome://settings/help. For administrators, the useful check is the version on managed devices, not the assumption that the rollout has completed, because Google says the release is staged over days and weeks.
Reports also show that Chrome 157 introduced a Windows dynamic patching mechanism that applies security updates to newly started child processes, with a DynamicPatchingEnabled policy available to administrators. Its account described the feature as distinct from Edge’s planned seamless update and did not reconcile it with the general relaunch advice. The report is a single secondary description, so it does not remove the need to confirm the fixed build is running.
Restricted detail and the AI credit
Google has restricted some of what the advisory reveals. It says access to bug details and links may be restricted until a majority of users have updated, and may remain restricted where a flaw exists in a third-party library that other projects depend on but have not fixed. The public record therefore gives components and bug classes without full technical paths for every flaw. No threat actor or campaign has been identified. No affected-user count or operational impact is established, and the status of proof-of-concept code is unknown.
The advisory also carries an AI-related credit that should be read narrowly. Google credits Xinyang Ge of Anthropic, assisted by Claude, for the critical Prerender use-after-free and for high-severity flaws including CVE-2026-121185 and CVE-2026-121118. It credits OpenAI Codex Security (amyb) for CVE-2026-121166 and CVE-2026-121047. Google does not state the degree of AI involvement. The credits establish AI involvement in research that produced reports Google credited; they do not establish that a model discovered Chrome flaws autonomously.
Chrome 157 is therefore best understood as a preventive release with a large fix count and a small group of severe flaws. The practical response is to confirm the fixed build and relaunch, not to treat the release as evidence of compromise. The open question is whether any of the seven critical flaws will later be shown to have been exploited. For now, the documented step is to ensure that the updated browser is the one actually running.
Seven critical ratings in one release is a good moment to ask what a severity score actually measures, and what it leaves out. The 9.6 base score attached to one flaw in this batch describes assessed severity, contributed by CISA-ADP, not an attack in progress, and the advisory reports no exploitation. Learning to hold those two things apart sits within internet vulnerabilities: how weaknesses in web-facing software are described, graded and disclosed, and why running a fixed build is a separate question from installing one. If you would like that grounding before the next advisory lands, explore The Hack Academy’s online cybersecurity training programme, where internet vulnerabilities is one of the listed subject areas and study is self-paced, with practical exercises set alongside the theory.