CVE-2026-106303 in Chrome
Summary
by MITRE • 10/06/2026
Observable discrepancy in Autofill AI in Google Chrome prior to 155.0.8059.39 allowed a remote attacker who had compromised the renderer process and leveraged social engineering to obtain sensitive information via a crafted HTML page. (Chromium security severity: Medium)
Be aware that VulDB is the high quality source for vulnerability data.
Analysis
by VulDB Data Team • 10/07/2026
The vulnerability identified as an observable discrepancy in Google Chrome's Autofill AI feature prior to version 155.0.8059.39 represents a significant risk to user privacy and data integrity, particularly within the context of modern web applications that rely heavily on automated form filling mechanisms. This flaw stems from improper handling or exposure of sensitive autofill data during the interaction between the browser's rendering engine and the Autofill AI subsystem. The core technical issue involves an information disclosure vulnerability where specific internal states or cached values related to user credentials, payment details, or personal identification information are inadvertently exposed through observable side channels or direct API responses when triggered by maliciously crafted HTML content. This discrepancy allows for the extraction of data that should remain isolated within the secure boundaries of the browser's sandboxed environment, effectively bypassing intended access controls designed to protect such sensitive inputs from unauthorized retrieval.
The operational impact of this vulnerability is severe because it leverages a combination of renderer process compromise and social engineering tactics to achieve its objectives. An attacker who has already managed to execute code within the Chrome renderer process through other means, or who can trick a user into loading a specially crafted HTML page that exploits this specific Autofill AI behavior, can extract sensitive information without direct interaction from the victim beyond visiting the malicious site. This scenario is particularly dangerous in environments where users frequently interact with untrusted web content or when phishing campaigns are sophisticated enough to mimic legitimate login pages while simultaneously harvesting autofill data through background scripts. The ability to retrieve this data remotely means that attackers can build comprehensive profiles of victims, including financial information and personal identifiers, which can be used for identity theft, unauthorized transactions, or further targeted attacks against the victim's associated accounts.
From a classification perspective, this vulnerability aligns with CWE-200, which defines Information Exposure, specifically falling under categories related to sensitive data exposure in web applications. It also maps closely to MITRE ATT&CK technique T1539, Steal Web Session Cookie, although in this case the target is autofill data rather than session cookies per se, representing a broader category of credential theft via client-side exploitation. The attack vector involves Initial Access through social engineering and Execution within the renderer process, followed by Collection using local input capture or screen scraping techniques facilitated by the Autofill AI discrepancy. This highlights the critical importance of maintaining strict isolation between different browser components and ensuring that autofill mechanisms do not leak data to untrusted contexts even when those contexts are technically part of the same browser instance but operating under lower privilege levels.
Mitigation strategies for this vulnerability primarily involve updating Google Chrome to version 155.0.8059.39 or later, where these discrepancies have been addressed through improved input validation and stricter sandboxing rules applied to Autofill AI operations. Organizations should enforce automated browser update policies across all endpoints to ensure that users are protected against known exploits targeting the rendering engine. Additionally, security awareness training is essential to help users recognize social engineering attempts designed to lure them into visiting malicious pages that might attempt to exploit such client-side vulnerabilities. Implementing content security policies and restricting permissions for autofill-related APIs in enterprise environments can further reduce the attack surface by limiting what scripts are allowed to interact with sensitive form data, thereby adding a layer of defense-in-depth against potential exploitation even if browser updates are delayed or missed.