An unreleased frontier large language model from Anthropic has autonomously converted a 9.8-severity use-after-free (UAF) vulnerability in the Linux kernel — CVE-2026-98432 — which had previously remained purely theoretical, into a fully weaponized exploit chain capable of achieving remote root access. The flaw resides in the IPv4 TCP handshake process and has existed since Linux 4.4, meaning it has been embedded in mainstream server kernels for nearly a decade. Exploitation requires no credentials, is network-reachable, and needs no user interaction; any server with MPTCP listening enabled is vulnerable. Because Go 1.24 and later versions enable MPTCP by default, a large number of related services are now at risk of exposure. As of now, major enterprise distributions such as Ubuntu and Red Hat have yet to release official patches, and only a handful of distributions have completed kernel-level fixes. The core takeaway from this incident is that AI has compressed the time required to weaponize a vulnerability from weeks — or even months — down to a matter of hours, dismantling the time-gap buffer that previously shielded defenders. This dramatically lowers the bar for advanced cyberattacks and represents a game-changing shift in the cybersecurity landscape. Probing traffic tied to this vulnerability has already been observed in the wild. Organizations can protect themselves by applying patches, temporarily disabling MPTCP, filtering TCP packets with abnormal IP options, and taking inventory of related assets.