Incident report · Intel · Microsoft · Samsung

A preinstalled Intel driver is silently breaking IPv6, OneNote and Microsoft 365

Intel Connectivity Performance Suite rewrites the IPv6 Flow Label to zero partway through every TCP connection. Microsoft's edge routes on that label, so the rest of each connection lands on a server that never saw it start, and gets reset. OneNote stopped syncing for nearly three weeks, and the cause was nowhere a normal user or support desk would look.

Fixed by Intel, still shipped broken by OEMs via Windows Update Reproduced 2026-10-07 Workaround available Updated 2026-10-28

Microsoft's engineers confirmed it the next day. Twenty days later, nothing has changed

On 2026-10-08, the day after receiving the packet-level evidence, Microsoft's Azure networking engineering confirmed the mechanism in writing and said it is working on the Azure Front Door side to reduce the impact, and with the Windows team on a fix. It has given no date. Microsoft's executive customer-relations team then named no engineering owner and no date for stopping Windows Update from offering the broken builds, and pointed the evidence to Feedback Hub, which it says "does not guarantee an individual response, an engineering owner, or a remediation date". The root-cause reports had been sitting in Feedback Hub since 2026-10-02. On 2026-10-20 Microsoft's Update Catalog still listed all three builds reported broken, with the same entry counts as on 2026-10-07. Intel acknowledged in writing, citing the owner's tests, that its own generic build does not have the bug, called the OEM package Samsung's responsibility, and published no advisory; on 2026-10-26 it closed the support case, writing that it had passed the findings to "the appropriate OEM teams as well as Microsoft", without saying which driver version fixed the bug. Samsung has not answered: it forwarded the complaint about a US-model laptop bought in the US to Samsung Russia, apparently because the owner's email address ends in .ru, and Samsung Russia replied on 2026-10-09 that it could not help and advised contacting the Samsung office of the country where the laptop was purchased. That is the US, where the complaint started. Full timeline · Status of every report.

Windows Update shipped the broken builds, and Intel's fix isn't reaching people

On 2026-10-07 the failure was reproduced on demand, and Intel's own generic ICPS 50.26.623.243 was tested on the same laptop: it does not zero the flow label. The broken OEM build reached that laptop through Windows Update, three times in 2026. Microsoft's Update Catalog lists the builds reported as broken under Samsung, Lenovo, VAIO, ASUS, HP, Dynabook and Huaqin. Of 1,086 ICPS entries from 19 companies, only one is the fixed build. Intel's release notes don't mention the fix, and there is no advisory. Who ships the affected builds · The October 7 test.

Intel was told in September 2025, and closed the thread twice

An engineer with an HP laptop posted this exact mechanism on Intel's own forum more than a year ago: the Intel connectivity service clears the IPv6 flow label and breaks cloud load balancing. Intel closed the inquiry for lack of a response, the engineer re-posted in November, and Intel closed it again. Organizations running HP EliteBooks have since reported Microsoft 365 and SharePoint resets in the same thread. One says Windows Update reinstalled the software after they removed it. Similar Killer-era reports go back to 2021. Read the prior reports.

IPv6 connections OK, OEM build

138 / 170

34 sites, ICPS 40.25.926.173 running (Oct 7)

IPv6 connections OK, Intel generic build

169 / 170

same laptop, ICPS 50.26.623.243 (Oct 7)

Packets with the flow label zeroed

1,385 → 0

OEM build vs. generic build

OneNote desktop sync

~3 weeks

broken, with empty notebooks shown as "Up to date"

Affected? Check in one command

In PowerShell: 1..10 | % { curl.exe -6 -sS -o NUL -w "%{http_code}`n" https://api.onedrive.com/ }. Any "Connection was reset" lines, on a PC with Intel Connectivity Performance Suite, point to this bug. Check, fix and undo instructions.

What is going on

The driver is IntcCo11X64.sys 11.5.11.19, the "Intel Connectivity Traffic Control Callout Driver" (service INTCCoSvc, WFP provider "Rivet Networks, LLC - RFE version 6.1.0.1"). It ships as part of Intel Connectivity Performance Suite 40.25.926.0 (app version; its driver packages are 40.25.926.173), preinstalled on the affected Samsung Galaxy Book. The Windows stack sends the SYN with a proper IPv6 Flow Label. The Intel callout then re-emits every ACK and data segment with the label set to 0.

RFC 6437 expects the label to stay constant for the life of a flow. Microsoft's Azure Front Door hashes on it, so the label-0 packets go to a different node, which answers with a reset. The SYN-ACK comes back at hop limit 47 and the RST at hop limit 49, so two different machines are answering. IPv4 has no flow label and is unaffected. That is why OneNote on the web and on a phone kept working and nobody could explain the failure. Full wire evidence and A/B table.

Three companies, three failures

Intel

Shipped a kernel driver that breaks a basic IPv6 rule on every connection, then fixed it silently: no advisory, nothing in the release notes. The broken builds keep reaching laptops through OEM packages, and its forum told users to uninstall.

Samsung (and other OEMs)

Preinstalled it and published the broken build on Windows Update. Samsung's one fixed submission targets other device IDs. Lenovo, VAIO, ASUS, HP and Dynabook publish broken builds too. Samsung's US Office of the President has not answered the complaint; Samsung forwarded it to Samsung Russia (apparently because of the owner's .ru email address), which sent it back to "the country where the laptop was purchased": the US.

Microsoft

Signed the broken driver and delivered it through Windows Update. Its own edge then falls over on it, where Google, Cloudflare and Akamai don't. OneNote shows the dead connection as "Up to date", and an escalated support case opened in August never found the cause. Its network engineers confirmed the mechanism within a day of seeing the evidence; its executive customer-relations team named no owner and no date for a fix.

Read the specific asks for each vendor.

How it was found

Sign-out, cache resets, Office repair and reinstalling are all irrelevant to this defect, and none of them could have found it. In this case, a Microsoft support escalation ran for five weeks without finding it, and an OpenAI Codex session gave up on it. Anthropic's Claude, running as an agent on the affected machine, then found it in a few hours, without knowing about Intel's earlier thread. It timed the resets against the round trip, captured traffic below every filter driver and decoded the IPv6 headers. It inventoried every Windows Filtering Platform callout and disabled network filter drivers one at a time, reversibly, until only the Intel driver was left.

Reported to

Plus private Intel, Microsoft and Samsung support cases. Samsung's community board hid the Samsung report as spam until a moderator restored it. Status of every report.

Seeing this too?

Add your laptop model, ICPS version and Windows build to the tracking issue. Don't post IP addresses, email addresses or ticket numbers.