Key Takeaways
- Linux 核心漏洞 CVE-2026-98432(CVSS 9.8)可能讓攻擊者無須登入,即可遠端取得 Root 權限。 CVE-2026-98432 是 Linux 核心 IPv4 TCP 程式碼中的釋放後使用(Use-After-Free,UAF)漏洞。攻擊者可能透過特定時序的 TCP SYN 封包觸發漏洞,無須經過應用程式或登入驗證。Red Hat 亦將此漏洞評為 9.8,Amazon Linux 則評為 8.1,顯示不同廠商的風險評估有所差異。
- Anthropic AI 研究模型已成功開發出可遠端取得 Linux Root 權限的漏洞攻擊程式。 Anthropic 指出,其尚未公開的 Claude 研究模型成功開發出漏洞攻擊程式,可讓未經身分驗證的遠端攻擊者,在啟用多路徑 TCP(Multipath TCP,MPTCP)監聽服務的伺服器上取得 Root 權限。不過,Anthropic 尚未公開攻擊程式碼與測試使用的 Linux 核心組態,這項研究成果也不代表漏洞已遭廣泛利用。
- CVE-2026-98432 尚未列入 CISA KEV,但已有疑似針對受影響系統的掃描活動。 截至原文發布時,此漏洞尚未被美國網路安全暨基礎設施安全局(CISA)納入已知遭利用漏洞清單(KEV),但已有疑似威脅行為者在網際網路上掃描受影響系統的通報。雖然尚無法據此確認成功入侵案例,企業仍應優先盤點受影響的 Linux 系統並評估曝險風險。
- Linux 修補進度因發行版本而異,企業應優先更新核心並強化防護。 截至原文發布時,Red Hat Enterprise Linux(RHEL)已提供修補程式,Ubuntu 26.10 也已完成修補,但部分 Ubuntu LTS 版本的更新仍處於測試階段。企業應依各發行版本的安全公告確認修補狀態,並注意停用 MPTCP 並非完整的緩解措施。使用 TrendAI Vision One™ TippingPoint™ 的企業,應確認已啟用 2026 年 11 月 3 日發布的過濾器 48774。
What the bug is
When a Linux server accepts a TCP connection, the kernel hands IPv4 options from the pending request to the new socket without copying them. If the socket is destroyed while another CPU is still building a SYN-ACK from the same options, the kernel reads memory that has already been freed. The trigger is SYN requests carrying the IPv4 Record Route or Timestamp options, followed by a transmission timed to hit that window. A second path opens when request-socket migration is enabled (net.ipv4.tcp_migrate_req=1 or a BPF reuseport program that migrates requests).
The upstream fix delays freeing the memory until every reader is finished (kfree_rcu). IPv6 copies options and avoids the main race, though a related IPv6 issue in the migration path, affecting options that only CALIPSO sets, still needs a separate patch.
Fixed upstream kernels include: 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10 and 7.3. Distribution backports keep their own version numbers, so we recommend checking vendor advisories rather than version strings.
Red Hat shipped fixed kernels for RHEL 8, 9 and 10 on November 1, and a reboot is required after installing them. Red Hat says that disabling MPTCP is not a complete mitigation and that default RHEL, which ships with MPTCP off, is still vulnerable until patched. Ubuntu’s 26.10 is fixed, fixes for 22.04, 24.04 and 26.04 LTS are in the -proposed pocket pending qualification testing and are not yet available through default updates, and fixes for older releases are still pending.
Why it matters
Anthropic said an AI model produced a working remote kernel exploit , though the exploit has not been released, and the bug class (a race in core networking code reachable before authentication) is thewhose exploitation has historically been left to a few specialist teams, according to Anthropic. The open questions are how exposed real systems are, since Anthropic's exploit targets servers with an MPTCP listener (though the underlying bug is not specific to MPTCP), and how fast others reproduce the result. Both questions decide whether this stays a research milestone or becomes an incident.
On exposure, this exploit targets servers with an MPTCP listener, though the underlying bug is not specific to MPTCP. MPTCP defaults vary: Red Hat documents it as off by default on RHEL, while it is reportedly on for recent Debian and Ubuntu installs. Go 1.24 and later enable it on Linux listening sockets by default. As of Nov. 4, neither Red Hat nor Ubuntu had released a fix to their default update channels; Ubuntu has fixes in, that are proposed for 22.04 and newer.
There have been reports of activity that appears to be a threat actor scanning the internet for affected systems.
What organizations should do
- Inventory internet-facing Linux servers and track your distribution's advisory for CVE-2026-98432; deploy fixed kernels as they ship and reboot (or apply a vendor live patch where one is offered).
- Check whether MPTCP listeners are enabled on exposed services, and whether net.ipv4.tcp_migrate_req is on.Turning off what you do not need is a temporary measure only: it doesn’t remove the vulnerability, so patching is still required.
- When you run TrendAI Vision One™ TippingPoint™, confirm filter 48774 is enabled.
- Treat the distribution scores as a reason to plan, not to wait: the severity disagreement is about timing difficulty, and AI-assisted exploit development is exactly what reduces that.
How TrendAI™ can help
TrendAI™ TippingPoint™ filter 48774, published on November 3, 2026, covers CVE-2026-98432 at the network level, so TippingPoint™ customers can block exploitation attempts, if any, while kernels are unpatched.