By Hamza Faisal
The Xen Project has patched two critical flaws in its hypervisor that could let a single malicious virtual machine read and write the memory of every other virtual machine on the same host — including the privileged management domain that controls the whole system. The bugs, disclosed on November 3 in advisory XSA-523, affect every Xen release from version 4.6 onward, and there is no workaround. The only fix is to install the patched build and reboot the host.
The flaws are tracked as CVE-2026-102703 and CVE-2026-102704, and both sit in Xen's grant-copy operation, the mechanism virtual machines use to pass data between one another through the hypervisor. In the advisory's words, a malicious guest can read and write arbitrary memory belonging to other domains, including Dom0 — in practice, a full escape from the virtual machine's isolation.
A broken isolation guarantee
The two bugs are logic errors in how Xen handles batches of copy operations. For performance, Xen caches the validated description of one copy and reuses it for the next element on the same page — but in some cases the conditions for reusing that cached check were wrong. CVE-2026-102703 lets frame numbers and grant references get mixed up, so a guest can reach memory it was never granted. CVE-2026-102704 is a misplaced permission check: copying by raw frame number is only allowed when a guest touches its own memory, but the check enforcing that rule sat in the wrong place.
The result is about as bad as a hypervisor flaw gets. A compromised guest could silently read secrets from neighbouring virtual machines or overwrite their memory, and because Dom0 itself is exposed, an attacker could take over the physical host. Cloud providers, hosting companies and anyone running multi-tenant Xen infrastructure are the most exposed, and downstream projects built on Xen — including XCP-ng and Qubes OS — have issued their own warnings and patches.
How Claude found the bugs
The disclosure is notable for who found the flaws. The Xen Project formally credits Anthropic, using its Claude AI model, with the discovery — the first time a commercial AI system has been publicly credited with finding critical flaws in an upstream hypervisor project. Stephan Lipp of Ada Logics validated the findings, and the CodeAnt AI Security Research Team reported the same issue independently. The patch commits even carry a trailing line reading "Assisted-by: Claude."
The credit lands in a week when AI-discovered flaws are making headlines: OpenAI said on November 4 that an unreleased cyber model built 948 working Linux kernel exploits in a single automated run. The message is the same on both fronts — autonomous agents are now finding serious security holes faster than most human teams.
Patch now — there is no workaround
Xen has released fixes for versions from xen-unstable through 4.19.x, as well as the 4.18.x and 4.17.x branches. Versions 4.6 through 4.16 have no patches in this advisory — support for 4.16 ended in December 2024 — so operators on those releases need to upgrade. XCP-ng has already shipped a patched build for its 8.3 release, and Qubes OS users should apply the project's update as soon as it lands.
There is no known exploitation of either flaw, and neither CVE appears in the US CISA Known Exploited Vulnerabilities catalog as of November 4. That is no reason to wait: the advisory is public, the mechanism is documented in detail, and the window before the first working exploit is usually short. Anyone running Xen should check their version today and schedule the reboot.



