Key Notes
- Anthropic says an exploit written by an unreleased Claude research model reached root-level command execution on a vulnerable Multipath TCP server in about 53 minutes.
- CVE-2026-98432 is rated 9.8 and has upstream fixes, but Ubuntu’s tracker still lists several standard kernel packages as awaiting corrections.
- The Go Security Team warns that the underlying bug is broader than Multipath TCP and recommends a patched kernel rather than relying on feature-level mitigations.
Anthropic says a Claude-written exploit achieved remote root-level command execution on a vulnerable Linux server in about 53 minutes, without credentials or user interaction. The demonstration described in its research report targeted a server exposing a Multipath TCP listener, an important condition behind the headline result.
The flaw, CVE-2026-98432, affects the Linux kernel’s IPv4 TCP handling and carries a 9.8 critical CVSS score in the Ubuntu advisory. Fixes are available upstream, but distribution packages are at different stages of rollout. For administrators, the immediate task is to establish whether their running kernel contains the correction.
What the 53-Minute Demonstration Shows
Anthropic reports roughly 48 minutes to obtain information needed for the attack, followed by another five minutes to execute a command as root. That is the runtime of its proof of concept, rather than a claim that Claude discovered the vulnerability and completed the entire research process in 53 minutes.
The work used an unreleased internal research model, with initial prompting from researchers. It demonstrates the model’s ability to produce a working exploit under those conditions; it does not identify an equivalent capability available to every user of the consumer Claude app.
The kernel announcement dates the CVE disclosure to October 15. Anthropic published its broader research report on October 27, and the Go Security Team issued a follow-up warning on October 29. Those are distinct stages in the disclosure and response, rather than a vulnerability first appearing today.
Multipath TCP Is a Key Condition, With a Wider Warning
The underlying problem is a race condition that can leave the networking code using memory after it has been freed. Root-level execution is especially serious because it gives an attacker privileges beyond a normal application account, potentially allowing changes to the host and access to sensitive information.
In its security notice, the Go team says Anthropic’s demonstrated exploit requires an exposed Multipath TCP listener. It also warns that the bug itself is not specific to that feature, so systems without such a listener could still be vulnerable to other attacks using the same underlying flaw.
The notice matters to Go developers because Go 1.24 and later enable Multipath TCP by default for Linux listen sockets. A service can therefore expose the relevant listener through its runtime defaults. The team describes disabling the feature program by program as a fragile mitigation and strongly recommends a patched kernel instead.
The Fix Must Reach the Kernel a Server Is Running
The Linux kernel team lists branch-specific corrected releases, including 6.12.113 and 7.2.10, and recommends updating to a supported stable kernel containing the fix. Distributions can backport security changes into their own package versions, so an upstream version number alone is not a reliable verdict on a vendor-supported installation.
As checked on October 30, Ubuntu’s tracker still marked the standard linux package for 22.04 LTS, 24.04 LTS, 26.04 LTS and 26.10 as vulnerable, with work in progress on the listed fixes. That makes “patched upstream” different from “already fixed in every distribution.” Administrators should consult the entry for their exact kernel package and release, since cloud and hardware-specific variants have separate rows.
Once a vendor’s correction is available, install the supported security update and verify that the fixed kernel is actually running, following that vendor’s activation or restart instructions. Where a package is still awaiting a fix, exposure and temporary mitigations need to be reviewed against current vendor guidance. Disabling one networking feature should not be treated as proof that the underlying kernel bug has disappeared.
Our coverage of Claude’s browser exploits examined another part of the same research program. The Linux case extends the concern to remotely reachable server infrastructure, where no employee has to open a document or visit a web page for the demonstrated attack to proceed.
Finding flaws is one part of the defensive response, alongside services such as Anthropic’s OSS Scanner. The operational challenge is turning a confirmed finding into an available vendor package and then into a fix running on the affected host. This case shows why each of those steps matters.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
