AI Science Tracker

Linux epoll upward-depth walk can access a freed eventpoll object

Linux kernel epoll · AI involvement attributed to Claude

Finding and attribution

An epoll nesting-depth walker follows an eventpoll pointer while under RCU, but the pointed-to eventpoll object is freed without waiting for an RCU grace period. Anthropic describes the resulting use-after-free and a suggested kfree_rcu fix.

Product: Linux kernel epoll. AI system: Claude. Team / organization: Anthropic.

CVE identifiers: CVE-2026-43074.

AI and human contribution (source account): The individual Anthropic CVD page explicitly credits Claude with discovery and Anthropic security with triage.

Human involvement is only described where the source account establishes it; unknown involvement does not mean none. Vendor confirmation of a flaw does not confirm every claim about AI contribution.

Discovery context: The CVE record was published 2026-05-06.

Remediation status: fix documented. The Anthropic report links the upstream kfree_rcu fix commit; Red Hat’s entry describes the same CVE and flaw but says Red Hat products are not affected.

A source-documented fix does not establish installed patches or current safety. Source checking does not independently reproduce the finding or establish observed exploitation.

Disclosed 6 May 2026. AI attribution dated 18 Aug 2026. Sources checked 15 Nov 2026; these are separate dates, not a live installation check.

Recorded CVSS: High · highest recorded score 7.8. Documented technical effects: other.

Severity, consequences and exploitation evidence

Case grouping uses the highest recorded CVSS score, with its version-specific severity band. It is descriptive, not a combined or editorial risk index. All recorded assessments remain below; scores from different versions and sources may disagree. CVSS specifications define version-specific scoring and severity bands.

  • CVE-2026-43074: CVSS 3.1 · 7.8 (High) · CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H · Assessment scenarios: AV:L - The vulnerable path is reached through local epoll syscalls, primarily `epoll_ctl(EPOLL_CTL_ADD)` on nested epoll descriptors racing with descriptor close/release. Network input is not sufficient to reach the lifetime bug without local control of epoll file descriptors. AC:L - The race is attacker-controlled: an unprivileged local process can create the epoll graph and run concurrent threads performing nested `epoll_ctl()` and close/release operations. No rare external state or victim timing is required beyond winning a race the attacker can repeatedly drive. PR:L - Creating epoll descriptors and calling `epoll_ctl()` require only ordinary local user privileges. No capability check gates the vulnerable nested epoll path; the only capability-related logic masks `EPOLLWAKEUP` and is not relevant. UI:N - Exploitation does not require another user to open files, mount anything, or perform any action. The attacking process can create, nest, and close all relevant descriptors itself. S:U - The impact is within the kernel security authority as a local kernel memory corruption or privilege escalation. There is no VM escape, IOMMU bypass, or cross-authority boundary change shown by this bug. C:H - This is a use-after-free of `struct eventpoll` during an RCU walk, and the stale pointer can be dereferenced after the object is freed and potentially reallocated. Per kernel CVSS guidance, a UAF is treated as enabling high confidentiality impact through exploitable memory corruption. I:H - The stale `struct eventpoll` is not only read but also written via fields such as `gen` and `loop_check_depth` after free, making heap corruption plausible if the memory is reallocated. Per kernel CVSS guidance, UAF memory corruption supports high integrity impact. A:H - At minimum, dereferencing or writing a freed `struct eventpoll` can crash or corrupt the kernel. A local attacker can repeatedly trigger the race, so availability impact is high.. Linux (CNA; CVE Program record) · checked 16 Nov 2026.

CVE-2026-43074: other.
Prerequisites / affected scope: Linux eventpoll may free struct eventpoll while another thread still uses it; the patch defers free through RCU. Downstream effect not established..
Exploitation evidence: The cited source does not establish exploitation in the wild.. Impact source · checked 16 Nov 2026.

Documented fix: The Anthropic report links the upstream kfree_rcu fix commit; Red Hat’s entry describes the same CVE and flaw but says Red Hat products are not affected.

Original evidence

Download this source record (JSON)

Open this case in the interactive tracker · Browse all static security cases