The ground underneath

Lets start in the kernel. A race condition in the Linux kernel’s TCP receive path, CVE-2026-99288, lets an unauthenticated attacker crash a server. Researcher Hyunwoo Kim published the advisory on November 15, after the linux-distros embargo ended, and the flaw is 11 years old — introduced by a commit from October 3, 2015, in Linux 4.4. They are calling it NoisyTCP.

The shape of it is a stale established-hash lookup racing the freeing and reallocation of a request socket in a different receive context, and the kernel dereferences a NULL listener. Panic. Any plain TCP listener is reachable, and the defaults are the targets — Nginx, Apache, SSH. It needs multi-CPU packet processing, two or more receive queues or Receive Packet Steering, and the attacker spreads the work with the receive-hash difference between IPv4 fragments and regular TCP packets. On QEMU/KVM virtual machines, they crash in about five seconds to two minutes. The fix landed upstream in 6.12.117, 6.18.63, and 7.3.4, but most major distros had not shipped fixed kernels as of November 15, and 5.10, 5.15, 6.1, and 6.6 had no fixed upstream release at all. The mitigations are to disable SYN cookies — at the cost of SYN-flood protection — or serialize to a single receive queue. Neither is free. NoisyTCP

Two things are worth getting right, because they will get confused. This is a denial of service, not remote code execution. Kim and the vendors document only the crash, and there is no exploitation in the wild and no CISA KEV entry. And it is not CVE-2026-98432, a separate TCP race rated 9.8 with remote root, disclosed by Anthropic researchers from the same introducing commit. Same commit, different flaw. The severity is unsettled in the ordinary way: the kernel CVE team and NVD had no CVSS at publication, Red Hat rates it 7.5, Oracle 8.1. The number matters less than the shape — eleven years of uptime, and the crash was already in the code.

The clock stops waiting

Google Project Zero announced on November 13 that it is replacing its “90 + 30” disclosure policy — 90 days to fix, plus 30 days of soak time for patch rollout, in place since 2021 — with “Full X”: a vulnerability is disclosed X days after it is reported to the vendor, whether or not a fix exists. Full 90 is effective immediately. Full 60 starts February 2, 2027. Full 30 starts January 3, 2028. The grace period and the variant policy are gone. What remains: vulnerabilities actively exploited in the wild, handled under Google’s application-security disclosure policy; already-public vulnerabilities, at Project Zero’s discretion; vendor-requested disclosure. Rare hardware exceptions, the Spectre and Meltdown kind, continue. Project Zero

The reasoning is stated plainly in the post. LLMs have increased the speed of attacker discovery and exploitation and reduced patch-to-exploit time to near zero — “zero patch-to-exploitation time is a reasonable, though imperfect, model.” Soak time is “no longer meaningful.” Report volume is up too: “several vulnerabilities a week in targets where we once found a few vulnerabilities a year.” And severity does not move the date: “A vulnerability’s severity does not affect the likelihood that it will be discovered by attackers, only its impact on users.” The phase-in was deliberate, after vendor feedback. An immediate jump to 30 days risked vendors accepting they could not meet the deadline and not attempting fixes at all.

Read it alongside NoisyTCP and the shape of the year comes into focus. The kernel flaw sat for eleven years, and the disclosure clock is now measured in weeks. The clock used to wait for the fix. Now the fix has to keep up with the clock.

The toolkit and the team

Elastic Security Labs published on November 12 its write-up of SANSAR, a Linux cryptomining toolkit tracked as intrusion REF3364. The orchestrator is written in Rust, version 0.1.3, takes commands over Tor through Arti, the Rust Tor implementation, and supervises CPU and GPU miners — plus shell and traffic tasks. The miners are managed, not written, by SANSAR: XMRig 6.25.0 on CPU, SRBMiner-MULTI 3.6.9 on GPU. Pool and wallet configuration arrive remotely. SANSAR

The evasion is the part that stays with you. A kernel rootkit, acpi_thermal_mgr.ko, hides selected processes, files, and socket entries. And a modified NVIDIA Management Library wrapper, replacing libnvidia-ml.so.1, falsifies the GPU readings — utilization reported as 0%, temperature randomized between 36 and 46°C, power between 100 and 200W. The card is mining and the dashboard says it is idle. There is a second backdoor, called netns, that gets its controller address from an Ethereum smart contract before connecting directly. Entry was an internet-exposed Ray job agent accepting unauthenticated job submissions. The timeline: external probes October 22, an attacker-submitted job ran nvidia-smi and exfiltrated the GPU inventory October 24, a Python reverse shell October 25. Persistence through SSH authorized keys, the toolkit staged under /opt/vadi_/. Three Linux GPU hosts compromised. Elastic did not disclose victim identities, crypto amounts, or financial cost, and found no evidence the operation extended beyond the three hosts.

The miner hides in the kernel. The dashboard lies about the GPU. The controller address lives on a blockchain.

Kaspersky ICS CERT published on November 5 its report on BO Team — also called Black Owl, Hoody Hyena, Lifting Zmiy — inside a Russian software developer whose products run in industrial plants, mostly for customers in Eastern Europe. The company contacted Kaspersky in June 2025 after suspicious activity, and Kaspersky held the report about a year so the victim could close the gaps. Attribution is “with a high degree of confidence”: BO Team, assessed to be of Ukrainian origin, active since at least 2023, focused on espionage and hacktivism against Russian entities. BO Team

Entry was an exposed SSH service on a Linux terminal server, then network reconnaissance. The attackers exploited Log4Shell, CVE-2021-44228, in VMware vCenter 7.0.3 to get a root shell — Kaspersky reproduced the exploit on the same vCenter build. Root on vCenter handed them its SSH private key, and from there the hypervisors. Persistence was hourly cron backdoors disguised as PostgreSQL and VMware components, timestomped to 2021, fetching scripts from free nofrillsdns subdomains parked on 127.0.0.1 until activated. Virtual machines got similar cron scripts plus a modified ReverseSSH backdoor. The attackers also trojanized the NX remote-desktop login script to harvest passwords, brute-forced root SSH on an admin server, and tried a WMI-delivered backdoor on the code-signing server, deleting logs throughout.

By the end they held nearly everything — the Windows domain, the hypervisors, GitLab, Ansible, backups, the mail server, the DNS zone — and had compromised the CI/CD infrastructure and the software code-signing server. The attack was detected and stopped before the objective was achieved. No malicious software release was confirmed, and whether code-signing certificates or private keys were exfiltrated could not be confirmed. A five-year-old flaw, a reproduced exploit, and the keys to the software supply chain.

The patch queue

VulnCheck disclosed on November 12 fourteen vulnerabilities in Copeland XWEB EVO, the Dixell refrigeration supervisor now sold under Copeland, CVE-2026-124002 through CVE-2026-124015. The NVD records went live at 20:18 UTC. The first exploitation attempts arrived about an hour and forty minutes later. VulnCheck’s canary sensors logged real attacks on CVE-2026-124014, an unauthenticated write of the system configuration file through config.lua — 87 attempts in the first hour from a single source. A second source began probing CVE-2026-124010, unauthenticated root command execution, at 03:41 UTC on November 13. Both went onto VulnCheck’s KEV list the same day. No public exploit code existed at the time. XWEB EVO

The worst of the fourteen is CVE-2026-124010, CVSS 9.8: one unauthenticated GET to the setuid-root CGI xwconsole.cgi runs the sh parameter as a root command. CVE-2026-124002, also 9.8, is operating-system command injection in the login page’s username field, before the password check. CVE-2026-124004, also 9.8, is a hardcoded administrator backdoor password. Eight of the fourteen are Critical, five High, one Medium. Copeland ended software support for the whole XWEB EVO family on March 16, 2026, after stopping sales in October 2023. These flaws will not be fixed. Shodan counts roughly 1,663 XWEB EVO systems on the internet; Onyphe counted about 1,960. And this is the same controller family CERT-FR named among the compromised systems in the October attack on roughly 700 French supermarkets, forced into defrost overnight — a campaign France attributed on November 9 to Iran’s IRGC cyber unit Shahid Kaveh. Disclosed on a Wednesday. Attacked within two hours. Patched never.

CVE-2026-129483, published November 15: Cyrus SASL through 2.1.28, SRP server mechanism only. In plugins/srp.c, the server checks only that the client’s public value A is greater than zero, and never rejects A where A modulo N equals zero — violating RFC 5054 and RFC 2945. Sending A equal to N forces the shared secret S to zero and the session key K to SHA1 of an empty string — a constant the attacker already knows. From there the attacker computes a valid M1 and authenticates as any existing user, without the password, and learns the session keys too. Reported by “afldl” in July 2026; the issue was published July 31. A full end-to-end proof of concept against the real libsasl2 is attached to it. The fix exists as a commit on master, but there is no fixed release — 2.1.28 is still the latest release, and there has not been a new one since 2022. The workaround is to disable the SRP mechanism. Not in CISA KEV; no known exploitation in the wild. One figure to carry carefully: the CVE record scores it CVSS 4.0 9.3, CWE-287. The reporter’s own issue scores it 9.8, CWE-347. The CVE record is the authoritative one. The secret was zero. The key was a constant. The password was never needed. Cyrus SASL

Dell’s DSA-2026-479, published November 13, covers multiple PowerStore vulnerabilities. The one that leads is CVE-2026-103813, CVSS 9.1 Critical: command injection where a low-privileged user runs a crafted command and gets a root shell — input passed directly to the operating system without sanitization. Behind it: CVE-2026-86356, 8.8, an authorization bypass to administrator; CVE-2026-82156, 8.8, another authorization bypass to administrator on the 500T; CVE-2026-107522, 7.2, improper authorization to administrator; CVE-2026-107520, 5.4, stored cross-site scripting; CVE-2026-82175, 3.8, server-side request forgery. No known exploit; not in CISA KEV. Dell PowerStore

Samsung’s November bulletin fixes ten Smart Switch flaws, two high and eight moderate. CVE-2026-21166, high: improper authentication in versions before 3.7.74.7 lets an adjacent attacker install arbitrary applications. CVE-2026-21167, high: Smart Switch binds to an unrestricted IP address, and an adjacent attacker can bypass authentication, with user interaction required. The rest include unauthorized APK restoration, APK verification bypass, and writing arbitrary files with Smart Switch privileges. Fixed in 3.7.74.7. No active exploitation stated. Samsung Smart Switch

And Octopus Server, CVE-2026-93487, CVSS 4.0 7.1 High: incorrect validation of the permissions for project triggers lets a user run runbooks or create deployments they were not authorized to perform. Advisory 2026-20, released November 9. The patch shipped September 28 — 2026.2.13487 — and there is no mitigation, only the upgrade. Found during internal testing by Nathan Willoughby. Three vendors, three different doors, one instruction: patch. Octopus Server

The incident ledger

Coles Group told the ASX on Monday, November 16 — disclosed just after 8:30 AM local, authorized by the board — that it is investigating a cybersecurity incident from over the weekend. The company’s own words: “Based on the findings of our investigation to date, we have not identified any evidence that customer, team or supplier data has been compromised or removed from Coles’ systems.” It is working with the Australian Cyber Security Centre and global cybersecurity experts to contain and secure the IT environment. Stores and online shopping are operating normally; the app and website are available; gift cards work. The only confirmed disruption is a small number of customers having problems with Liquorland click-and-collect orders. Coles

What is not disclosed: what kind of incident, when it was detected, which systems are involved, the attack vector, or any threat actor. No ransomware claim, and no actor has claimed responsibility. Coles is asking customers to watch for scam emails and texts, and has published an incident updates page and a customer response line. One thing not to conflate: Woolworths customers reporting unauthorized transactions on digital WISH gift cards, reported the same day, is a separate and unrelated item. No data loss found — so far. The words are careful, and the careful words are the news.

The November 10 Patch Tuesday security updates broke Windows activation and licensing operations. slmgr.vbs /ipk, which installs a product key, and slmgr.vbs /skms-domain, which points a machine at a KMS server, fail with error 0x80041008 — “Invalid parameter.” Microsoft confirmed it in service alert WI1549486. Five emergency out-of-band cumulative updates shipped on Friday, November 13: KB5136489 for Windows Server 2025, KB5136487 for Server 2022, KB5136490 for Server 2019 and Windows 10 Enterprise LTSC 2019, KB5136483 for Windows 11 23H2, and KB5136493 for Windows 10 22H2 ESU and Windows 10 Enterprise LTSC 2021. Microsoft activation

The scope is organizations using SLMGR and KMS automation: Windows 10 21H2 and 22H2, Windows 11 23H2, and Windows Server 2019, 2022, and 2025. Windows 11 24H2, 25H2, and 26H2 were not affected. Microsoft’s reassurance: devices that are already activated remain activated, and KMS-activated devices typically remain activated up to 180 days after successful activation. Distribution is mostly the Microsoft Update Catalog only; the Windows 11 23H2 fix is also offered through Windows Update under Optional updates. All five are cumulative and include the November 10 security fixes. A security update that broke the thing that proves the machine is licensed. Fixed in three days, out of band.

Tor 0.4.9.17 shipped November 13, announced by Tor Network Team developer David Goulet as an emergency release in “rapid response to the ongoing DDoS attack on the network, which has affected most relay operators.” The attack is a massive flood of client circuits overwhelming relays — a single relay could be hit with hundreds of circuits per second. The key fix, TROVE-2026-031: relays now track IPv6 clients by their /64 prefix instead of individual addresses for denial-of-service-defense counting. It is a bugfix on 0.3.3.2-alpha, bug 41344 — the flaw lived in the code for years. Also in the release: TROVE-2026-075, dropping DNS resolution replies with an invalid hostname; keys removed for bastet, a directory authority, over a potential compromise; the maatuska directory authority’s IP addresses changed; fallback directories regenerated November 13; geoip files updated. Tor

It follows 0.4.9.16 on November 5 and 0.4.9.15 less than three weeks earlier, which added three denial-of-service protections against a “thunderstorm” of directory requests. Tor urges relay operators to upgrade as soon as possible; no further action is required after upgrading. More releases are expected as additional defenses are tested. This is core relay software, not a Tor Browser build. What is not disclosed: no CVE number, no traffic volumes, no attacker attribution. Relay operators reported disruption continuing after the update — a November 14 forum post noted relays losing HSDir status. The fix counts attackers by the network they come from, not the address they use. The attack is still going.

Leave a Reply

Trending