Lightpanda CVE-2026-52843

Exploitation requires a Lightpanda session (version prior to 0.2.9) whose shared cookie jar already contains valid session cookies for a target/victim origin, and requires attacker-controlled web content to execute inside that same session and issue a fetch() or XMLHttpRequest to the victim origin - this is the UI:R interaction. The specific broken behavior is that credentials: omit, credentials: same-origin, and XMLHttpRequest.withCredentials=false are all ignored, so an attacker's cross-origin request that should carry no cookies is sent authenticated. Limiting factors: the affected version must be < 0.2.9; the session must be holding cookies for a sensitive origin at the time attacker content runs; and the attacker must get their content loaded into the automation session, so a pipeline that never mixes authenticated state with untrusted browsing is not exploitable.

The published CVSS 3.1 vector (AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:N, 9.3) is internally consistent with the described flaw: network-reachable, low-complexity, no privileges, but requiring interaction (the session must load attacker-controlled content), with a scope change because the attacker abuses one origin's context to act against a different victim origin, yielding high confidentiality and integrity impact and no availability impact. No EPSS score was provided, so widespread-exploitation probability cannot be quantified from the input. There is no KEV entry and no evidence of a public POC - only the vendor patch/commit is public - so this is a high-severity design flaw rather than a confirmed in-the-wild threat. The practical risk is bounded by deployment: Lightpanda is a specialized automation browser, so exposure depends entirely on whether a given automation pipeline browses untrusted origins while holding authenticated cookies for sensitive origins. SSVC-style read: exploitation is not observed, technical impact is total for the credential-leak primitive, and automatability is plausible since a single malicious page can drive it - but there is no independent confirmation of active exploitation.

An operator runs a Lightpanda automation session that is logged into an internal application (holding its session cookies), then directs or allows the session to load an attacker-controlled page. That page issues a fetch() or XMLHttpRequest to the internal application's origin with credentials: omit set; Lightpanda ignores the directive, attaches the victim's session cookies, and the attacker's script reads the authenticated response. No public POC is identified at time of analysis, and the AV:N/AC:L profile means the malicious page needs only to be loaded once (UI:R) to trigger the leak.