All Posts

The Justice Department unsealed an indictment this week against three Russian nationals and two companies for running the infrastructure under a big slice of the cybercrime economy. The companies are Media Land LLC and ML.Cloud LLC, both out of St. Petersburg. The people are Alexander Volosovik, Kirill Zatolokin, and Yulia Pankova. The charge sheet — computer-fraud conspiracy, wire fraud, money laundering — covers 42 victims across 21 US states and roughly 62 million dollars

Cisco Talos published a campaign today worth two things: a clear look at where command-and-control infrastructure is heading, and an honest test of our own feed that we failed. Both are worth your time, and we are going to give you both, because the second one is the reason to trust the first. First, a correction, because getting the category right is the job. This is not ransomware and it is not an extortion crew, whatever the shorthand says. Talos tracks the actor as UAT-11

Arctic Wolf Labs published a good piece of primary research this week. Their team documented a threat actor running at least 292 fake GitHub repositories, each impersonating a trusted brand, each a lure to install an in-memory infostealer. The list of impersonated brands spans security tooling, fintech, crypto wallets, developer tools, secure email, macOS utilities, and gaming — and in a detail you could not make up, one of the fake pages impersonated Arctic Wolf itself. The

Yesterday's AsyncAPI supply-chain attack pulled its second-stage payload — a RAT called Miasma — from IPFS, the distributed content-addressed file network. That is a deliberate choice, and it is becoming a common one. Attackers reach for IPFS because there is no single server to seize. You cannot send a takedown notice to a content hash. It reads as un-seizable, and that is exactly why we have been building against it since early July. We resolved yesterday's payload to the m

On July 14, an attacker published malware into the @asyncapi namespace on npm — five poisoned versions across four packages that pull more than two million downloads a week. The interesting part is not that it happened. Supply-chain compromises happen weekly, and we cover them weekly. The interesting part is what the attacker did not need to do, and where they hid the payload once they were in. On the second point, we have a receipt most write-ups do not, because we built the

Microsoft's July 14 Patch Tuesday closed 570 vulnerabilities — 59 of them Critical — and three zero-days. Two of those zero-days are being exploited in the wild right now. Here is the part that should reorganize your patch queue: neither of the two exploited bugs is a remote-code-execution flaw. Both are elevation of privilege. And both were found by incident responders picking through live breaches, not by researchers in a lab. One is CVE-2026-56164 in SharePoint, which we w

SonicWall confirmed this week that two flaws in its SMA1000 remote-access appliances have been exploited as zero-days, in the wild, chained together. CISA added both to the Known Exploited Vulnerabilities catalog on July 14 and set a hard clock: federal agencies have until July 17 under Binding Operational Directive 26-04 to patch the appliance or rip it out of service. Three days. Patch or unplug. There is no third option in the directive. We did not have these two CVEs pre-

On July 14, CISA did two things about Microsoft SharePoint on the same day. It added CVE-2026-56164 to the Known Exploited Vulnerabilities catalog, and it published a hardening advisory warning that on-premises SharePoint Server is under active attack. The advisory is the more important of the two, because it names not one flaw but three, and it says they are being used together. The three are CVE-2026-32201, CVE-2026-45659, and CVE-2026-56164. CISA added them to KEV on three

Matt Lucas at RedEye Security and Etairos published something today that deserves a wider frame than "neat computer-vision trick," because it is the newest entry in a lineage that is exactly forty years old. Mixfont's "ghost font" encodes text purely in motion: every single frame of the video is television static, nothing for OCR to grab, and yet a person watching it reads the words instantly. The pitch is that only human eyes can read it — screenshot it and the text is gone.

This morning we published JADEPUFFER — the first ransomware operation run end-to-end by an AI agent — and the single most useful line in that piece was not about the attack, it was about the defense: the AI-generated payload narrated its own reasoning, in prose, inside the code, and that self-narration is a detection surface that did not exist a year ago. We said the tell of AI-written attack code is that it cannot help but explain itself. That was published before lunch. By

On July 13 we published a post about UniFi with an unusually honest opening: we said, up front, that we did not have a receipt on this one — no prior flag, no confirmed in-the-wild exploitation, nothing to point at. It was a patch-ahead warning, not a victory lap. The argument was that the thing making UniFi dangerous is the exact thing that keeps it off everyone's threat model: people trust it because it looks clean, and the trust is the vulnerability. We wrote it anyway, wi

Wiz published GhostApproval on July 8, and it is the cleanest proof yet of a thing we started writing in April: the AI coding agent is now the login shell, and the security industry is still protecting the old one. The finding is not a bug in one tool — Wiz is explicit that it is a category-level design gap across six of the most popular AI coding assistants — and the mechanism is so old it is almost insulting. It abuses the symbolic link, a Unix feature older than most of th

For two years the phrase "AI-powered attack" has mostly meant a human criminal using a chatbot to write a better phishing email. That is not this. Security researchers at Sysdig have documented JADEPUFFER, and the reason it matters is not that AI helped — it is that no human was in the loop for the attack itself. A human set up the infrastructure and pointed the thing at a target. From there, an autonomous AI agent did the entire job: found the way in, exploited it, looked ar

Lidl told customers in Germany, Belgium, and the Netherlands this week that their personal data was stolen — and then said the sentence every breached brand now says: the online shop itself was not affected. The data lived in a separately stored customer database maintained by a third-party IT service provider, and that provider is where it walked out. Titles, first and last names, phone numbers, email addresses, dates of birth, and customer numbers. No passwords, no billing

ReliaQuest documented two new Microsoft 365 phishing kits this week, Jalisco and OmegaLord, and the interesting thing about them is not that they exist — phishing kits are a commodity — but that they attack multi-factor authentication from two completely different angles, and neither of them is the thing your MFA rollout was designed to stop. One of them never asks the victim for a password at all. Together they are a clean illustration of a shift we have been writing about f

Every major vendor wrote React2Shell up in December. Microsoft, Google, Wiz, Cloudflare, Rapid7, Zscaler, JFrog, Qualys, Sysdig — a wall of coverage the week it dropped, and then the news cycle did what it always does and moved on. Our exploit harvester did not move on, because it does not read the news; it watches GitHub. And what it has quietly recorded since is the part the disclosure-week coverage cannot show you: a 10.0 remote-code-execution bug does not get weaponized o

This is what the space between a disclosure and a catastrophe looks like when you watch it in real time. CVE-2026-49049 is an unauthenticated flaw in Helix3, the template framework by JoomShaper that quietly underpins a huge number of Joomla sites. Over the last nine days we watched public exploit tooling for it accumulate across four separate GitHub accounts, one of them an operation we already had a name for. And while that assembly line was filling up, the payload it produ

Let us be clear about what CVE-2026-11405 is, because the word "vulnerability" is doing a lot of quiet work to make it sound like an accident. CERT/CC disclosed it on July 6 as VU#213560, and the mechanism is not a buffer overflow, not a missing bounds check, not a mistake a tired developer makes at 2 a.m. It is a second password. It was put there on purpose. And the routers it lives in are sitting in hundreds of thousands of homes and small offices, unpatched, because the ve

Twelve hours ago we published an honest miss: the ChocoPoC campaign — fake proof-of-concept exploit repositories that pull one poisoned PyPI dependency to compromise the security researchers who run them — had slipped both our defenses, and we said so plainly. This post is what happened next, in one working day, and it ends with a live malicious account still serving payloads that our brand-new detector found and GitHub's takedown had missed. What we built between the two pos

If you are the kind of person who reads this blog, you are the target of this one. ChocoPoC is a malware campaign that does not go after enterprises or end users — it goes after the vulnerability researchers and pentesters who download proof-of-concept exploits from GitHub to test the week's hot CVE. That is our audience, and it is us. So this is written straight, with the honesty axis pointed at ourselves first: we checked our own feed for this campaign's indicators, and we