NoisyTCP Can Crash Linux Servers, Plus 4 Stories
The NoisyTCP Linux vulnerability is the lead today because it turns ordinary internet traffic into a way to crash an entire server before anybody logs in. The rest of the brief is also about access and limits: an AI attack tool moving faster than many security teams, Gemini reading financial history, Facebook reaching into phone galleries, and Microsoft repairing an activation problem that mostly affects managed Windows fleets.
NoisyTCP can crash a Linux host through an open port
An attacker doesn’t need an account on your website, a valid SSH login, or a WordPress vulnerability to exploit NoisyTCP, tracked as CVE-2026-99288. The vulnerable code lives in the Linux kernel’s TCP networking stack. If the attacker can send packets to an exposed service such as Nginx, Apache, SSH, or a proxy, carefully timed traffic can trigger a kernel panic and take down the host.
The mechanism is a race condition in the kernel’s lockless socket lookup. Linux is checking one connection while memory from an earlier request is being freed and reused for another. By spreading fragmented IPv4 and ordinary TCP packets across different receive queues, the attacker can make the kernel grab a reference to the replacement object before it’s fully attached to a listening socket. The kernel follows a null pointer and crashes.
That’s the technical explanation. The practical one is simpler: the web server application may be configured correctly and still disappear because the operating system underneath it fell over. Containers don’t provide a separate kernel, so triggering the flaw against a containerized web service can crash the whole host and every other container sharing it.
The vulnerable code dates to Linux 4.4 in October 2015. Upstream fixes are in stable kernels 6.12.117, 6.18.63, and 7.3.4, but don’t use those numbers alone to decide whether a production server is safe. Ubuntu, Debian, Red Hat, cloud providers, and appliance vendors backport fixes into their own kernel packages. Check the advisory for the distribution and release you’re actually running.
So far, the public research demonstrates a remote denial of service, not code execution or data theft. The researcher reproduced VM crashes in roughly five seconds to two minutes, while a tuned attack against a quiet bare-metal server took about 14 hours. I haven’t seen evidence of exploitation in the wild, but the proof of concept and technical detail are public now. This is patch-and-reboot work, not panic-and-rebuild work.
Install the vendor’s fixed kernel, schedule the reboot, and verify the running kernel afterward. Downloading a package doesn’t replace the kernel already in memory. Temporary workarounds exist, including disabling SYN cookies or reducing the external interface to one receive queue, but those trade away SYN-flood protection or network performance. Don’t make that bargain casually just because the command is shorter than a maintenance window.
Business IT goes well beyond the website
Your business also depends on workstations, cloud accounts, browsers, Wi-Fi, remote access, collaboration tools, and all the other technology that quietly becomes infrastructure. Raymond Tec works across that whole stack, whether the problem lives on a server, on a desk, or somewhere in between.
ARTEX makes familiar attack paths move much faster
The FBI is warning about ARTEX, an autonomous penetration-testing framework that uses large language models and specialized agents to handle reconnaissance, exploitation, credential theft, cloud access, persistence, and data removal. It was designed for authorized security testing. Attackers have been using it against real organizations since at least August.
The important claim isn’t that AI discovered a magical new class of vulnerability. ARTEX lets an operator provide a goal and have software chain familiar techniques across multiple targets without waiting for a person to drive every step. The FBI says some intrusions moved from initial targeting to privileged access in under three hours. That matters when a business still reviews alerts the next morning, or when a service provider needs half a day to decide whether an unusual login is worth calling about.
The alert connects ARTEX to attacks against South Korean financial institutions and to a ConnectWise Cloud Backup supply-chain incident affecting more than 2,250 downstream tenants. It also says researchers found more than 1,600 publicly exposed ARTEX consoles in early November. That number doesn’t mean all 1,600 belonged to criminals. It does mean the framework kept spreading after the original repository disappeared, through forks, archives, localized versions, and a commercial edition.
For a small business, the useful response is not “buy AI security.” Patch internet-facing applications quickly, require phishing-resistant multifactor authentication where possible, limit service accounts to the data and actions they need, and make sure somebody can review cloud identity activity, unusual downloads, and new persistence within hours rather than days. If an outside provider manages Microsoft 365 or remote support, ask how quickly it receives and acts on those alerts. As last week’s ScreenConnect campaign showed, legitimate administrative reach is exactly what makes these systems attractive.
Technical discovery & auditing
The public page doesn’t tell you much about the machinery behind it. Raymond Tec audits inherited and long-running projects to uncover the plugins, integrations, data, dependencies, and old decisions that determine what the next change will really involve.
Gemini can read Google Wallet and linked bank activity
Google is rolling out an optional Google Wallet connection for Gemini to all Gemini app users in the United States. After enabling it in Connected Apps, a user can ask Gemini to summarize spending, find recurring subscriptions, compare price increases, or build a budget. Google Wallet activity is available directly, and users can optionally connect bank and credit-card accounts through Plaid for a broader view.
Google says the connection is read-only. Gemini can’t move money, manage an account, or make an investment through it. That’s a meaningful boundary. Read-only access is still access to a detailed record of where you shop, which doctors or charities you pay, when you travel, and which subscriptions you may prefer not to explain to a human reviewer.
That last phrase isn’t hypothetical. Google’s Connected Apps documentation says a subset of connected-app data is reviewed by trained human reviewers for debugging, safety, and improvement, and explicitly warns against connecting apps containing confidential information a user wouldn’t want a reviewer to see. Disconnecting Wallet later doesn’t delete information already stored in Gemini Apps Activity. That has to be managed separately, and material already reviewed may be retained longer.
I can see the usefulness here. Finding three forgotten subscriptions could pay for the inconvenience quickly. But don’t enable financial access because the toggle is new. Decide whether the questions you expect to ask justify the data you are granting, review Gemini’s activity and auto-delete settings first, and verify important totals against the actual account statements. A budgeting assistant can be useful without becoming the accounting system of record.
Italy says one Android permission wasn’t specific consent
Italy’s privacy regulator has fined Meta €805 million over Facebook’s “Get suggestions from gallery when browsing Facebook” feature on Android. The authority says Facebook used one broad app permission to access photos, videos, and metadata for several functions, while the gallery-suggestion feature was enabled by default and wasn’t explained clearly enough for informed consent. About 20 million Italian users were affected.
Meta says the penalty is disproportionate and plans to appeal. The ruling requires a separate, specific consent request and says the feature must be off by default within 60 days. It did not decide whether Meta used the gallery material to train AI, which Meta denies; the regulator says a newer cloud-processing feature may be examined separately.
The order applies in Italy, but the practical question travels well: did you give Facebook access to photos for one task and quietly authorize several others? Android users can review Facebook under Settings, Apps, Facebook, Permissions, then check the app’s own media and gallery-suggestion settings. Removing photo access may disable convenient upload or suggestion features. That is the tradeoff, and it should be the user’s tradeoff to make.
The rules around technology matter too
Platforms, privacy, speech, competition, surveillance, copyright, and regulation increasingly determine what technology companies can build and what the rest of us have to live with. Browse more Raymond Tec News for practical coverage of technology policy and digital rights.
Microsoft’s emergency activation fix is for a narrow problem
Microsoft has released out-of-band updates after November’s security patches caused some Software Licensing Management Tool commands to fail with error 0x80041008. The problem mainly affects organizations using SLMGR scripts to install product keys or configure Key Management Service activation. PCs that are already activated remain activated, and ordinary users don’t need to hunt for an emergency patch just because the headline says “Windows activation.”
Affected administrators should match the update to the platform: KB5136483 for Windows 11 23H2, KB5136493 for supported Windows 10 ESU and LTSC systems, KB5136490 for Windows Server 2019, KB5136487 for Server 2022, and KB5136489 for Server 2025. Server and most Windows 10 packages are available through the Microsoft Update Catalog; the Windows 11 fix is also an optional Windows Update. If provisioning works and no affected SLMGR command is failing, test normally instead of treating “out of band” as “install everywhere before breakfast.”
Still in a reading mood? The Raymond Tec News archive covers security, AI, small-business technology, policy, and the places technology collides with ordinary life — without requiring a computer-science degree to get through it.
Sources / Further Reading
- Hyunwoo Kim: NoisyTCP technical advisory and mitigations
- Linux kernel CVE team: CVE-2026-99288 affected and fixed versions
- Red Hat: CVE-2026-99288 impact and product status
- Debian Security Tracker: CVE-2026-99288
- BleepingComputer: NoisyTCP Linux server impact
- FBI/IC3: Threat Overview of the ARTEX Autonomous Penetration Testing Framework
- Check Point Research: ARTEX operator-console investigation
- Google Threat Intelligence: ARTEX and related offensive frameworks
- Google: Google Wallet integration for Gemini
- Google: Personalization and data handling with Connected Apps
- Google: Gemini Apps Privacy Hub
- Italian Data Protection Authority: Meta gallery-access decision
- Reuters: Meta’s response to the Italian privacy fine
- BleepingComputer: affected Windows platforms and out-of-band updates
- Microsoft: KB5136483 activation fix
Photo by Kevin Ache on Unsplash.
