Linux kernel vulnerabilities posted to a public mailing list: what ZDI sent, the two CVSS 9.0 ksmbd flaws, and whether AI reports are the cause

Linux kernel vulnerabilities reached public mailing lists without patches on October 21, 2026, when Trend Micro's Zero Day Initiative (ZDI) sent detailed bug reports, including proof-of-concept code, to the kernel security team with public lists copied. Two of the five kernel reports describe remote code execution in ksmbd, the kernel's built-in SMB file server, each rated CVSS 9.0 by ZDI. Kernel maintainers objected on October 22. The episode lands five months after Linus Torvalds said a flood of AI-generated reports had made the kernel's private security list "almost entirely unmanageable", and after the kernel changed its rules on how AI-found bugs should be reported.

This story is based on ZDI's reports and the maintainers' replies as archived on the public Linux kernel mailing list, the kernel's security-bugs documentation, Trend Micro's own pages, and reporting by Phoronix, The Hacker News, LWN and BetaNews, checked October 23, 2026. For how a different AI-found flaw was handled in October 2026, see our PHP2Shell story.

The short version

• What was posted: five kernel bug reports from ZDI, sent to the kernel security address with public lists copied, plus a sixth for ConnMan, a separate userspace network daemon.
• Most serious: two ksmbd remote code execution flaws, both CVSS 9.0. ksmbd is not Samba, and most systems don't run it.
• Patches: none in the mainline kernel as of October 23. The ksmbd maintainer says one ksmbd bug was already fixed in his tree, and a fix for the other was posted on October 23. No CVE IDs yet.
• The AI question: the kernel tells reporters to treat AI-found bugs as public. Nothing shows AI found these. The reports credit named researchers "working with TrendAI Zero Day Initiative," which is the program's current name, not a tool.

What was posted, and how serious is it?

The ratings below are ZDI's own, from its reports.

Last verified: October 23, 2026

A sixth report, ZDI-CAN-32005, describes a stack buffer overflow in ConnMan's handling of search-domain queries. ConnMan is a userspace daemon, not part of the kernel.

Fixes are moving. For the ksmbd dispatch race (ZDI-CAN-33774), ksmbd maintainer Namjae Jeon replied on October 22 that "this issue has already been fixed," linking a commit in his own Git tree. That commit credits a different reporter and was committed on October 20, the day before ZDI's report; it wasn't in the mainline kernel as of October 23. For the SMB Direct bug (ZDI-CAN-33549), SMB Direct maintainer Stefan Metzmacher asked the reporter on October 21 to test a branch aimed at Linux 7.5, and on October 23 Jeon posted a patch written by Metzmacher and marked for the stable kernels. For skb_zerocopy, networking developer Eric Dumazet replied that developers had received "a report with a patch from another source one week ago."

Who is exposed?

Fewer systems than the 9.0 ratings suggest. The ksmbd bugs matter only where the in-kernel SMB server runs. The Hacker News cites Red Hat's documentation that Red Hat Enterprise Linux doesn't ship ksmbd; administrators can check with lsmod | grep ksmbd. Cyber Kendra reported that ksmbd is off by default upstream and most common on home NAS boxes and single-board computers. The vmwgfx bug affects only VMware guests, and the nvmet bug only machines exporting storage over NVMe-oF.

The kernel's CVE team assigns IDs only after a fix reaches a stable tree, The Hacker News notes, so none of these bugs had a CVE on October 23. No exploitation had been reported, but the archived reports include proof-of-concept code and crash traces, so the technical details are public.

Was it a mistake?

Phoronix's headline called the reports "accidentally posted." Kernel maintainers treated them as a process failure. Greg Kroah-Hartman, the stable kernel maintainer, replied to three of them on October 22. On the vmwgfx report he wrote: "For some reason you sent this to public lists, and did not include a patch, so there's not much we can do with this." Less than a minute later, on the NVMe report: "Again, no patch, and sent to a public list? Please fix your submission process." And on the ksmbd race: "And again, no patch, and a public list?" Networking developer Willem de Bruijn pointed ZDI to the kernel's security-bugs document.

ZDI doesn't describe it as a mistake. In a reply on the list on October 22, it said it has "probably submitted nearly 1000 cases to kernel over the last ten years," formatted much as they always were, and that "we will send the cases to you as we always have." It said it won't supply patches ("it's overhead we cannot support") and will release its findings "at rejection, or 120 days." It didn't say why public lists were copied. Willy Tarreau wrote in the same thread that ZDI now copies the relevant maintainers, as the kernel had asked, and that when public lists are copied, the kernel's documentation says the private security address isn't needed. The thread turned heated: ZDI objected to one maintainer's insult as "racially insensitive."

Do the kernel's AI rules explain it?

Possibly in part, but nothing confirms that. The kernel's security-bugs document now has a section on AI-assisted reports.

The document, as published for kernel 7.3.0 and checked October 23, 2026, says: "If you resorted to AI assistance to identify a bug, you must treat it as public." Its reasoning is that the security team's experience shows such bugs "systematically surface simultaneously across multiple researchers, often on the same day." The same paragraph says not to publicly share a reproducer, only to mention that one exists. It also asks AI-assisted reporters to propose and test a fix first.

Torvalds explained the change in his Linux 7.1-rc4 announcement on May 17, 2026: "The continued flood of AI reports has basically made the security list almost entirely unmanageable," as quoted by BetaNews. On May 20 he added, per LWN: "if you found it with AI, a hundred other people also found it with AI."

ZDI's reports fit parts of that guidance and break others. Copying public lists is what the rules ask for AI-found bugs. But the reports came without patches, which is what Kroah-Hartman objected to, and they include full proof-of-concept code, which the rules say not to share publicly. ZDI's own account, that its reports are formatted as they have been for a decade, suggests its normal process rather than the AI rule.

Were these reports generated by AI?

Nothing in the public record says so. The Hacker News and Cyber Kendra wrote that the reports credit "TrendAI," which they called Trend Micro's AI scanner. The reports themselves, as archived, credit named researchers: the two ksmbd bugs to Hyeontae Lee and the skb_zerocopy bug to p1r0x, each "working with TrendAI Zero Day Initiative." That is the program's name. Trend Micro's ZDI page is titled "Zero Day Initiative (ZDI) | TrendAI" and describes a bug bounty program that pays independent researchers; it doesn't mention AI tools finding bugs. Trend Micro's site says "TrendAI™ is the enterprise security business of Trend Micro," and a Trend Micro partner notice says the enterprise business was renamed TrendAI effective March 25, 2026.

Whether the researchers used AI tools isn't known. ZDI's reply mentions AI only to say: "We understand that your workflow has increased because of AI... and we are in the same boat." Tarreau wrote in the same thread that "most reports these days likely come from AI tools," and Phoronix ended its story by noting that reports "keep on coming in the AI fueled world." Both remarks describe the kernel's workload in general, not these reports.

AI bug-finding at scale is real elsewhere: researchers at Calif used OpenAI's GPT-5.6 Sol to find PHP2Shell, the test case for OpenAI's Patch the Planet disclosure proposal. Maintainers use AI too, as in Torvalds' Claude Code page cache experiment, and projects are writing rules for it, as in our story on projects banning AI code.

What's confirmed, reported and uncertain

• Confirmed (Linux kernel mailing list archive): the reports reached the public list on October 21; ZDI's ratings for the ksmbd and skb_zerocopy bugs; the researcher credits; the replies from Kroah-Hartman and other developers; ZDI's reply; the ksmbd and SMB Direct fixes as their maintainers describe them.
• Confirmed (kernel docs; Trend Micro): the kernel's rules on AI-found bugs; TrendAI as the name of Trend Micro's enterprise business and of its ZDI program.
• Reported (The Hacker News, Phoronix, Cyber Kendra): the vmwgfx and nvmet ratings, the ConnMan report, and the description of TrendAI as an AI scanner, which the reports' credits don't support.
• Uncertain: whether the researchers used AI tools; why ZDI copied public lists; when fixes reach mainline and stable kernels, and when CVE IDs are assigned.