A vulnerability broker accidentally published six unpatched Linux kernel bugs to a public mailing list. GitLab shipped its second CVSS 10.0 patch in nine days, covering every self-managed release since 2014. And the UK announced a £600 million bet that AI itself is the next security frontier.
ZDI accidentally posts six unpatched Linux kernel flaws to a public mailing list
The standard order of vulnerability disclosure broke this week. Over October 20–21, Trend Micro’s Zero Day Initiative (ZDI) sent six vulnerability reports — five for the Linux kernel, one for the ConnMan network manager — to the public Linux kernel mailing list (LKML) and related lists, instead of keeping them on the private security address until fixes existed.
The batch includes two remote code execution flaws in KSMBD, the in-kernel SMB server, each rated CVSS 9.0:
| Report ID | Component | Bug | Impact |
|---|---|---|---|
| ZDI-CAN-33774 | KSMBD | Race condition in handler table assignment | RCE, no authentication, any server with KSMBD on port 445 |
| ZDI-CAN-33549 | KSMBD / SMB Direct | Integer underflow, crafted RDMA buffer descriptors | RCE; needs RDMA, a writable share, and SMB Direct |
| ZDI-CAN-33367 | skb_zerocopy (networking) | Page-cache write from a read-only file | Local privilege escalation, reproduced to a root shell |
| ZDI-CAN-34068 | vmwgfx (VMware GPU driver) | Out-of-bounds write | Local privilege escalation |
| ZDI-CAN-34076 | nvmet (NVMe target driver) | Use of uninitialized resource | Information disclosure |
| ZDI-CAN-32005 | ConnMan | Stack buffer overflow | Privilege escalation, posted a day earlier |
Several of the reports went out with full proof-of-concept source. None of the six had a CVE assigned or a merged mainline fix at the time they were posted. Stable kernel maintainer Greg Kroah-Hartman replied to several threads questioning why they were sent to public lists without patches — which does not un-send anything. Phoronix spotted the posts first on October 22, and The Hacker News has since confirmed the two critical KSMBD flaws.
One of the more unsettling details: TuxCare reproduced the skb_zerocopy bug (ZDI-CAN-33367) to a root shell on AlmaLinux, CloudLinux, Debian, and Ubuntu, and reports that it bypasses May’s fix for Dirty Frag (CVE-2026-43284). A patch was posted to the netdev list on October 22 and has been reviewed but not yet merged; KernelCare shipped a rebootless patch for EL10 on October 23.
Why it matters: attackers now have a head start on these bugs before patches reach distribution kernels. The mitigating facts: KSMBD is off by default and stays off unless an admin deliberately enables it, and the SMB Direct flaw has narrow requirements. If you run KSMBD anywhere, disable it if unused; otherwise install distro kernels as they ship. The finder credited on several reports is TrendAI, Trend Micro’s AI vulnerability scanner — a theme that runs through today’s other stories.
GitLab patches a second CVSS 10.0 flaw in nine days — this one hits 12 years of releases
On October 21, GitLab released critical patch versions for a flaw that lets an attacker run code on a self-managed GitLab server without logging in. CVE-2026-110264 is an integer overflow during compilation of a specially crafted regular expression in the bundled regex engine. GitLab scores it the maximum, CVSS 10.0.
The affected range is remarkable: every self-managed GitLab CE and EE release from 7.0 (released in 2014) through the current branch. This is the second CVSS 10.0 regex-engine bug in nine days — the October 12 releases fixed a separate flaw (CVE-2026-97229) in the same engine, and the new fixed versions sit one patch above those, so servers patched nine days ago are vulnerable again.
| If you run | Upgrade to |
|---|---|
| 19.5 before 19.5.1 | 19.5.1 |
| 19.4 before 19.4.3 | 19.4.3 |
| 19.3 before 19.3.6 | 19.3.6 |
| 19.2 before 19.2.9 | 19.2.9 |
| 19.1 before 19.1.9 | 19.1.9 |
| 19.0 before 19.0.10 | 19.0.10 |
| 7.0 through 18.11, before 18.11.13 | 18.11.13 |
There is no workaround; upgrading is the fix. Only self-managed installations need to act: GitLab.com already runs the patched version, and GitLab Dedicated tenants on recent internal releases already have the fix. GitLab has reported no sign the flaw has been exploited, and it is not in CISA’s Known Exploited Vulnerabilities catalog as of October 22. The unauthenticated attack path, per GitLab’s own detection rules, needs a public project holding a crafted Org-mode file that GitLab renders — narrowing the attack, but not closing it.
Why it matters: a self-managed GitLab server holds source code, CI/CD secrets, and deploy credentials — it is close to the worst single server an attacker can own. The “patch twice in nine days” pattern is exactly the failure mode that punishes anyone running a quarterly upgrade cadence.
AI is now finding vulnerabilities faster than humans can patch them
A Reuters report this week pulls the thread connecting the PHP, Linux, and ZDI stories: AI agents are discovering software vulnerabilities at a pace that is compressing the whole disclosure-and-patch cycle. Seven cybersecurity professionals told Reuters that AI-driven discovery is almost certain to trigger more disruption of the kind that took government websites across Europe offline this week — several simply pulled their sites down until they could patch.
The numbers and quotes that stand out:
- Research due to be published by threat-intelligence firm VulnCheck finds the number of software flaws discovered roughly doubling year-on-year, driven by AI.
- Cloudflare CEO Matthew Prince told Reuters he is aware of “a bunch of additional severe vulnerabilities that are present in other pieces of critical software” and predicted: “I think we’re going to have a couple years where every single week it’s going to feel like the sky is falling.”
- The US National Security Agency told organizations to patch PHP immediately or pull vulnerable systems offline — a call echoed by authorities from Japan to the European Union.
PHP itself disclosed a new informational advisory on October 23: CVE-2026-115560, a potential remote code execution flaw that requires an unusual configuration — the session.serialize_handler setting flipped between php_binary/php_serialize and php mid-request via ini_set() before session_start(), plus the session upload progress feature enabled. PHP considers that combination unlikely in typical applications and did not release an emergency patch, recommending the session feature be disabled until a regular release. If you are still on an unpatched PHP, the priority remains the PHP2Shell emergency pair: 8.2.35, 8.3.36, 8.4.27, or 8.5.12.
Why it matters: the kernel-security documentation notes that bugs found by AI scanners are treated as effectively public, because the same tools are widely available and surface the same issues simultaneously. Disclosure norms assumed a human-speed finder; the ZDI leak and the PHP saga show what happens when that assumption fails.
The UK puts £600 million behind AI security
On Saturday, the British government set out around £600 million (about $800 million) in funding aimed at defending against health and security threats from artificial intelligence. The largest share funds a new biosecurity initiative: £100 million for an AI laboratory to detect biological threats and £65 million for countermeasures against pathogens.
OpenAI and Anthropic will contribute up to £200 million in AI model credits to the biosecurity initiative, with Google DeepMind and the London life-sciences startup Basecamp Research offering other assistance. The package also gives the existing AI Security Institute an extra £120 million and £130 million for cyber defences for critical infrastructure through new programs including the National Cyber Shield and the National Exposure and Real-Time Vulnerability Centre. Further lines include £60 million for British companies using AI in robotics and manufacturing, £60 million for AI research and procurement, £18 million for a chip design academy, and £15 million for a prize for AI solutions to public policy problems. AI Minister Kanishka Narayan said the AI Biosecurity Lab is intended to become “a major global institution” against emerging biological threats.
Why it matters: this is one of the largest government packages explicitly framed as AI-safety spending rather than AI-capability spending — and it lands in the same week that AI-discovered vulnerabilities started taking government websites offline. The sequencing writes its own argument.
Key takeaways
- Check whether you run KSMBD on any Linux host with port 445 reachable; disable it if unused, and keep distro kernels current.
- If you self-manage GitLab, you are vulnerable to CVE-2026-110264 unless you upgraded this week — including instances patched on October 12. Upgrade to the fixed release for your branch.
- Plan for AI-driven vulnerability discovery as the new normal: shorter disclosure windows, more emergency patches, and a premium on fast patching cadence.
- PHP operators: the emergency pair (8.2.35 / 8.3.36 / 8.4.27 / 8.5.12) is still the priority; the new CVE-2026-115560 only affects an unusual session configuration.
References
- Hardware Busters: Linux Kernel Vulnerabilities Hit the Public Mailing List Before the Fixes, Two of Them Rated CVSS 9.0
- SecurityOnline: KSMBD and Linux Kernel Vulnerabilities Details Go Public
- TuxCare: Linux Kernel skb_zerocopy Zero Day Gets Root on AlmaLinux, Debian and Ubuntu
- The Hacker News: Critical GitLab Flaw (CVSS 10.0) Could Allow Unauthenticated RCE Across 12 Years of Releases
- Rescana: GitLab CVE-2026-110264 Critical Unauthenticated RCE
- Reuters via The Hindu BusinessLine: AI-discovered software vulnerabilities raise risk of cyberattacks and outages
- Analytics Insight: New PHP Security Flaw Raises Concerns After Emergency Patches (CVE-2026-115560)
- CNA: UK government sets out $800 million of AI security funding
Was this useful?
Thanks for the feedback.
