PhantomRaven Attack Exploits NPM’s Unchecked HTTP URL Dependency Feature

An example of RDD in a package’s dependencies list. It’s not even counted as a ‘real’ dependency. (Credit: Koi.ai)
Having another security threat emanating from Node.js’ Node Package Manager (NPM) feels like a weekly event at this point, but this newly discovered one is among the more refined. It exploits not only the remote dynamic dependencies (RDD) ‘feature’ in NPM, but also uses the increased occurrence of LLM-generated non-existent package names to its advantage. Called ‘slopsquatting’, it’s only the first step in this attack that the researchers over at [Koi] stumbled over by accident.

Calling it the PhantomRaven attack for that cool vibe, they found that it had started in August of 2025, with some malicious packages detected and removed by NPM, but eighty subsequent packages evaded detection. A property of these packages is that in their dependencies list they use RDD to download malicious code from a HTTP URL. It was this traffic to the same HTTP domain that tipped off the researchers.

For some incomprehensible reason, allowing these HTTP URLs as package dependency is an integral part of the RDD feature. Since the malicious URL is not found in the code itself, it will slip by security scanners, nor is the download cached, giving the attackers significantly more control. This fake dependency is run automatically, without user interaction or notification that it has now begun to scan the filesystem for credentials and anything else of use.

The names of the fake packages were also chosen specifically to match incomplete package names that an LLM might spit out, such as unused-import instead of the full package name of eslint-plugin-unused-imports as example. This serves to highlight why you should not only strictly validate direct dependencies, but also their dependencies. As for why RDD is even a thing, this is something that NPM will hopefully explain soon.

Top image: North American Common Raven (Corvus corax principalis) in flight at Muir Beach in Northern California (Credit: Copetersen, Wikimedia)

hackaday.com/2025/10/30/phanto…

Iconic Xbox Prototype Brought to Life

When Microsoft decided they wanted to get into the game console market, they were faced with a problem. Everyone knew them as a company that developed computer software, and there was a concern that consumers wouldn’t understand that their new Xbox console was a separate product from their software division. To make sure they got the message though, Microsoft decided to show off a prototype that nobody could mistake for a desktop computer.

The giant gleaming X that shared the stage with Bill Gates and Seamus Blackley at the 2000 Game Developers Conference became the stuff of legend. We now know the machine wasn’t actually a working Xbox, but at the time, it generated enormous buzz. But could it have been a functional console? That’s what [Tito] of Macho Nacho Productions wanted to find out — and the results are nothing short of spectacular.

The key to this project is the enclosure itself, but this is no simple project box we’re talking about here. Milled from a solid block of aluminum, the original prototype’s shell reportedly cost Microsoft $18,000 to have produced, which would be around $36,000 when adjusted for inflation. Luckily, the state of the art has moved forward a bit in the intervening two decades. So after working with [Wesk] to create a 3D model from reference images (including some that [Tito] took himself of one of the surviving prototypes on display in New York), the design was sent away to PCBWay for production. It still cost the better part of $6 K to be produced, but that’s a hell of a savings compared to the original. Though [Tito] still had to polish the aluminum himself to recreate the original’s mirror-like shine.

To say the rest of the project was “easy” would be something of an understatement, but it was at least more familiar territory. Unlike the original prototype, this machine would actually play Xbox games, to [Tito] focused on cramming the original era-appropriate hardware (plus a few modern homebrew tweaks, such as HDMI-out) into the hollow X using a clever system of integrated rails and 3D printed mounts.

Some of the original parts, like the power supply, were simply too large to use. That’s where [Redherring32] came in. He designed a custom USB-C power supply that could satisfy the original console’s energy needs in a much smaller footprint. There’s also a modern SSD in place of the 8 GB of spinning rust that the console shipped with back in 2001. But overall, it’s still real Xbox hardware — no emulation or other funny tricks here.

At this point, the team had already exceeded what Microsoft pulled off in 2000, but they weren’t done yet. Wanting to really set this project apart, [Tito] decided to replace the center jewel with something a bit more modern. The original was little more than a backlit piece of plastic, but on this build it’s a circular LCD driven by a Raspberry Pi Pico, capable of showing a number of custom full-motion animations thanks to the efforts of [StuckPixel].

The end result of this team effort is a machine that’s not only better looking than Microsoft’s original, but also more functional. It’s a project that’s destined for a more than just sitting on a shelf collecting dust, so we’re happy to hear that [Tito] plans on taking it on a tour of different gaming events to give the public a chance to see it in person. He’s even had a custom crate made so he can transport it around in style and safety.

youtube.com/embed/0OMP8JvGWNY?…

hackaday.com/2025/10/30/iconic…

Build Your Own Force-Feedback Joystick

Force feedback joysticks are prized for creating a more realistic experience when used with software like flight sims. Sadly, you can’t say the same thing about using them with mech games, because mechs aren’t real. In any case, [zeroshot] whipped up their own stick from scratch for that added dose of realistic feedback in-game.

[zeroshot] designed a simple gimbal to allow the stick to move in two axes, relying primarily on 3D-printed components combined with a smattering of off-the-shelf bearings. For force feedback, an Arduino Micro uses via TMC2208 stepper drivers to control a pair of stepper motors, which can apply force to the stick in each axis via belt-driven pulleys. Meanwhile, the joystick’s position on each axis is tracked via magnetic encoders. The Arduino feeds this data to an attached computer by acting as a USB HID device.

We’ve seen some other great advanced joystick projects over years, too. Never underestimate how much a little haptic feedback can add to immersion.

youtube.com/embed/YdNP5jIJ0dU?…

hackaday.com/2025/10/30/build-…

Why You Shouldn’t Trade Walter Cronkite for an LLM

Has anyone noticed that news stories have gotten shorter and pithier over the past few decades, sometimes seeming like summaries of what you used to peruse? In spite of that, huge numbers of people are relying on large language model (LLM) “AI” tools to get their news in the form of summaries. According to a study by the BBC and European Broadcasting Union, 47% of people find news summaries helpful. Over a third of Britons say they trust LLM summaries, and they probably ought not to, according to the beeb and co.

It’s a problem we’ve discussed before: as OpenAI researchers themselves admit, hallucinations are unavoidable. This more recent BBC-led study took a microscope to LLM summaries in particular, to find out how often and how badly they were tainted by hallucination.

Not all of those errors were considered a big deal, but in 20% of cases (on average) there were “major issues”–though that’s more-or-less independent of which model was being used. If there’s good news here, it’s that those numbers are better than they were when the beeb last performed this exercise earlier in the year. The whole report is worth reading if you’re a toaster-lover interested in the state of the art. (Especially if you want to see if this human-produced summary works better than an LLM-derived one.) If you’re a luddite, by contrast, you can rest easy that your instincts not to trust clanks remains reasonable… for now.

Either way, for the moment, it might be best to restrict the LLM to game dialog, and leave the news to totally-trustworthy humans who never err.

hackaday.com/2025/10/30/why-yo…

100-Year Old Wagon Wheel Becomes Dynamometer

If you want to dyno test your tuner car, you can probably find a couple of good facilities in any nearby major city. If you want to do similar testing at a smaller scale, though, you might find it’s easier to build your own rig, like [Lou] did.

[Lou’s] dynamometer is every bit a DIY project, relying on a 100-year-old wagon wheel as the flywheel installed in a simple frame cobbled together from 6×6 timber beams. As you might imagine, a rusty old wagon wheel probably wouldn’t be in great condition, and that was entirely true here. [Lou] put in the work to balance it up with some added weights, before measuring its inertia with a simple falling weight test. The wheel is driven via a chain with a 7:1 gear reduction to avoid spinning it too quickly. Logging the data is a unit from BlackBoxDyno, which uses hall effect sensors to measure engine RPM and flywheel RPM. With this data and a simple calibration, it’s possible to calculate the torque and horsepower of a small engine hooked up to the flywheel.

Few of us are bench testing our lawnmowers for the ultimate performance, but if you are, a build like this could really come in handy. We’ve seen other dyno builds before, too. Video after the break.

youtube.com/embed/61-e-HK6HdU?…

hackaday.com/2025/10/30/100-ye…

Half-good new Danish Chat Control proposal

Denmark, currently presiding over the EU Council, proposes a major change to the much-criticised EU chat control proposal to search all private chats for suspicious content, even at the cost of destroying secure end-to-end encryption: Instead of mandating the general monitoring of private chats (“detection orders”), the searches would remain voluntary for providers to implement or not, as is the status quo. The presidency circulated a discussion paper with EU country representatives today, aiming to gather countries’ views on the updated (softened) proposal. The previous Chat Control proposal had even lost the support of Denmark’s own government.

“The new approach is a triumph for the digital freedom movement and a major leap forward when it comes to saving our fundamental right to confidentiality of our digital correspondence”, comments Patrick Breyer (Pirate Party), a former Member of the European Parliament and digital freedom fighter. “It would protect secure encryption and thus keep our smartphones safe. However, three fundamental problems remain unsolved:

1) Mass surveillance: Even where voluntarily implemented by communications service providers such as currently Meta, Microsoft or Google, chat control is still totally untargeted and results in indiscriminate mass surveillance of all private messages on these services. According to the EU Commission, about 75% of the millions of private chats, photos and videos leaked every year by the industry’s unreliable chat control algorithms are not criminally relevant and place our intimate communication in unsafe hands where it doesn’t belong. A former judge of the European Court of Justice, Ninon Colneric (p. 34-35), and the European Data Protection Supervisor (par. 11) have warned that this indiscriminate monitoring violates fundamental rights even when implemented at providers’ discretion, and a lawsuit against the practice is already pending in Germany.

The European Parliament proposes a different approach: allowing for court orders mandating the targeted scanning of communications, limited to persons or groups connected to child sexual abuse. The Danish proposal lacks this targeting of suspects.

2) Digital house arrest: According to Article 6, users under 16 would no longer be able to install commonplace apps from app stores to “protect them from grooming”, including messenger apps such as WhatsApp, Snapchat, Telegram or Twitter, social media apps such as Instagram, TikTok or Facebook, games such as FIFA, Minecraft, GTA, Call of Duty, and Roblox, dating apps, video conferencing apps such as Zoom, Skype, and FaceTime. This minimum age would be easy to circumvent and would disempower as well as isolate teens instead of making them stronger.

3) Anonymous communications ban: According to Article 4 (3), users would no longer be able to set up anonymous e-mail or messenger accounts or chat anonymously as they would need to present an ID or their face, making them identifiable and risking data leaks. This would inhibit, for instance, sensitive chats related to sexuality, anonymous media communications with sources (e.g. whistleblowers), and political activity.

All things considered, the new Danish proposal represents major progress in terms of keeping us safe online, but it requires substantially more work. However, the proposal likely already goes too far already for the hardliner majority of EU governments and the EU Commission, whose positions are so extreme that they will rather let down victims altogether than accept a proportionate, court-proof and politically viable approach.”

patrick-breyer.de/en/half-good…

The Time Of Year For Things That Go Bump In The Night

Each year around the end of October we feature plenty of Halloween-related projects, usually involving plastic skeletons and LED lights, or other fun tech for decorations to amuse kids. It’s a highly commercialised festival of pretend horrors which our society is content to wallow in, but beyond the plastic ghosts and skeletons there’s both a history and a subculture of the supernatural and the paranormal which has its own technological quirks. We’re strictly in the realm of the science here at Hackaday so we’re not going to take you ghost hunting, but there’s still an interesting journey to be made through it all.

Today: Fun For Kids. Back Then: Serious Business

English churches abound with marble-carved symbols of death.
Halloween as we know it has its roots in All Hallows Eve, or the day before the remembrance festivals of All Saint’s Day and All Soul’s Day in European Christianity. Though it has adopted a Christian dressing, its many trappings are thought to have their origin in pagan traditions such as for those of us where this is being written, the Gaelic Samhain (pronounced something like “sow-ain”). The boundary between living and dead was thought to be particularly porous at this time of year, hence all the ghosts and other trappings of the season you’ll see today.

Growing up in a small English village as I did, is to be surrounded by the remnants of ancient belief. They survive from an earlier time hundreds of years ago when they were seen as very real indeed, as playground rhymes at the village school or hushed superstitions such as that it would be bad luck to walk around the churchyard in an anticlockwise manner.

As a small child they formed part of the thrills and mild terrors of discovering the world around me, but of course decades later when it was my job to mow the grass and trim the overhanging branches in the same churchyard it mattered little which direction I piloted the Billy Goat. I was definitely surrounded by the mortal remains of a millennium’s worth of my neighbours, but I never had any feeling that they were anything but at peace.

Some Unexplained Phenomena Are Just That

A previously unexplained phenomenon in the appropriately named Death Valley. Jon Sullivan, Public domain.
So as you might expect, nothing has persuaded me to believe in ghosts. I can and have walked through an ancient churchyard at night as I grew up next to it, and never had so much as a creepy feeling. I do however believe in unexplained phenomena, but before you throw a book at your computer I mean it in the exact terms given: observable phenomena we know occur, but can’t immediately explain.

To illustrate, a good example of a believable unexplained phenomenon was those moving rocks in an American desert; they moved but nobody could explain how they did it. It’s now thought to be due to the formation of ice underneath them in certain meteorological circumstances, so that’s one that’s no longer unexplained.

As another slightly less cut-and-dried example there are enough credible reports of marsh lights to believe that they could exist, but the best explanation we have, of spontaneous combustion of high concentrations of organic decomposition products, remains for now a theory. I hope one day a scientist researching fenland ecosystems captures one on their instruments by chance, and we can at last confirm or deny it.
The ghost hunting kit of 1920s paranormal investigator Harry Price. Harry Price, Public domain.
The trouble is with unexplained phenomena, that there are folks who would prefer to explain them in their own way because that’s what they want to believe. “I want to believe” is the slogan from the X Files TV show for exactly that reason.

People who want a marsh light or the sounds made by an old house as it settles under thermal contraction at night to be made by a ghost, are going to look for ghosts, and will clutch at anything which helps them “prove” their theories. In this they have naturally enlisted the help of technology, and thus there are all manner of gizmos taken into cemeteries or decaying mansions in the service of the paranormal. And of course in this we have the chance for some fun searching the web for electronic devices.

All The Fun Of Scam Devices

In researching this it’s been fascinating to see a progression of paranormal detection equipment over the decades, following the technological trends of the day. From early 20th century kits that resembled those used by detectives, to remote film cameras like the underwater Kodak Instamatic from a 1970s Nessie hunt we featured earlier this year, to modern multispectral imaging devices, with so much equipment thrown at the problem you’d expect at least one of them to have found something!
I coulda found GHOSTS with this thing, had I only thought of it!
I’ve found that these instruments can be broadly divided into two camps: “normal” devices pressed into ghost-hunting service such as thermal cameras or audio recorders, and “special” instruments produced for the purpose. The results from either source may be digitally processed to “reveal” information, much in the manner of the famous “dead salmon paper“, which used an MRI of a dead fish to make a sarcastic comment about some research methodologies.

I’ve even discovered that I may have inadvertently reviewed one a few years ago, a super-cheap electric field meter touted as helping prevent some medical conditions, which I found to be mostly useful for detecting cables in my walls. Surprisingly I found it to be well engineered and in principle doing what it was supposed to for such an instrument, but completely uncalibrated and fitted with an alarm that denounced the mildest of fields as lethal. At least it was a lot cheaper than an e-meter.

Tomorrow night, there will be those who put on vampire costumes to be shepherded around their neighbourhoods in search of candy, and somewhere in the quiet country churchyard of an Oxfordshire village, something will stir. Is it a spectre, taking advantage of their yearly opportunity for a sojourn in the land of the living? No, it’s a solitary fox, hoping to find some prey under the moonlight in the undergrowth dividing the churchyard from a neighbouring field.

Wherever you are, may your Halloween be a quiet and only moderately scary one.

Header: Godstone, Surrey: Gravestone with skull and bones by Dr Neil Clifton, CC BY-SA 2.0.

hackaday.com/2025/10/30/the-ti…

Self-Driving Cars and the Fight Over the Necessity of Lidar

If you haven’t lived underneath a rock for the past decade or so, you will have seen a lot of arguing in the media by prominent figures and their respective fanbases about what the right sensor package is for autonomous vehicles, or ‘self-driving cars’ in popular parlance. As the task here is to effectively replicate what is achieved by the human Mark 1 eyeball and associated processing hardware in the evolutionary layers of patched-together wetware (‘human brain’), it might seem tempting to think that a bunch of modern RGB cameras and a zippy computer system could do the same vision task quite easily.

This is where reality throws a couple of curveballs. Although RGB cameras lack the evolutionary glitches like an inverted image sensor and a big dead spot where the optical nerve punches through said sensor layer, it turns out that the preprocessing performed in the retina, the processing in the visual cortex and analysis in the rest of the brain is really quite good at detecting objects, no doubt helped by millions of years of only those who managed to not get eaten by predators procreating in significant numbers.

Hence the solution of sticking something like a Lidar scanner on a car makes a lot of sense. Not only does this provide advanced details on one’s surroundings, but also isn’t bothered by rain and fog the way an RGB camera is. Having more and better quality information makes subsequent processing easier and more effective, or so it would seem.

Computer Vision Things

A Waymo Jaguar I-Pace car in San Francisco. (Credit: Dllu, Wikimedia)
Giving machines the ability to see and recognize objects has been a dream for many decades, and the subject of nearly an infinite number of science-fiction works. For us humans this ability is developed over the course of our development from a newborn with a still developing visual cortex, to a young adult who by then has hopefully learned how to identify objects in their environment, including details like which objects are edible and which are not.

As it turns out, just the first part of that challenge is pretty hard, with interpreting a scene as captured by a camera subject to many possible algorithms that seek to extract edges, infer connections based on various hints as well as the distance to said object and whether it’s moving or not. All just to answer the basic question of which objects exist in a scene, and what they are currently doing.

Approaches to object detection can be subdivided into conventional and neural network approaches, with methods employing convolutional neural networks (CNNs) being the most prevalent these days. These CNNs are typically trained with a dataset that is relevant to the objects that will be encountered, such as while navigating in traffic. This is what is used for autonomous cars today by companies like Waymo and Tesla, and is why they need to have both access to a large dataset of traffic videos to train with, as well as a large collection of employees who watch said videos in order to tag as many objects as possible. Once tagged and bundled, these videos then become CNN training data sets.

This raises the question of how accurate this approach is. With purely RGB camera images as input, the answer appears to be ‘sorta’. Although only considered to be a Class 2 autonomous system according to the SAE’s 0-5 rating system, Tesla vehicles with the Autopilot system installed failed to recognize hazards on multiple occasions, including the side of a white truck in 2016, a concrete barrier between a highway and an offramp in 2018, running a red light and rear-ending a fire truck in 2019.

This pattern continues year after year, with the Autopilot system failing to recognize hazards and engaging the brakes, including in so-called ‘Full-Self Driving’ (FSD) mode. In April of 2024, a motorcyclist was run over by a Tesla in FSD mode when the system failed to stop, but instead accelerated. This made it the second fatality involving FSD mode, with the mode now being called ‘FSD Supervised’.

Compared to the considerably less crash-prone Level 4 Waymo cars with their hard to miss sensor packages strapped to the car, one could conceivably make the case that perhaps just a couple of RGB cameras is not enough for reliable object detection, and that quite possibly blending of sensors is a more reliable method for object detection.

Which is not to say that Waymo cars are perfect, of course. In 2024 one Waymo car managed to hit a utility pole at low speeds during a pullover maneuver, when the car’s firmware incorrectly assessed its response to a situation where a ‘pole-like object’ was present, but without a hard edge between said pole and the road.

This gets us to the second issue with self-driving cars: taking the right decision when confronted with a new situation.

Acting On Perception

The Tesla Hardware 4 mainboard with its redundant custom SoCs. (Source: Autopilotreview.com)
Once you know what objects are in a scene, and merge this with the known state of the vehicle and, the next step for an autonomous vehicle is to decide what to do with this information. Although the tempting answer might be to also use ‘something with neural networks’ here, this has turned out to be a non-viable method. Back in 2018 Waymo created a recursive neural network (RNN) called ChauffeurNet which was trained on both real-life and synthetic driving data to have it effectively imitate human drivers.

The conclusion of this experiment was that while deep learning has a place here, you need to lean mostly on a solid body of rules that provides it with explicit reasoning that copes better with what is called the ‘long tail’ of possible situations, as you cannot put every conceivable situation in a data set.

This thus again turns out to be a place where human input and intelligence are required, as while an RNN or similar can be trained on an impressive data set, it will never be able to learn the reasons for why a decision was made in a training video, nor provide its own reasoning and make reasonable adaptations when faced with a new situation. This is where human experts have to define explicit rules, taking into account the known facts about the current surroundings and state of the vehicle.

Here is where having details like explicit distance information to an obstacle, its relative speed and dimensions, as well as room to divert to prevent a crash are not just nice to have. Adding sensors like radar and Lidar can provide solid data that an RGB camera plus CNN may also provide if you’re lucky, but also maybe not quite. When you’re talking about highway speeds and potentially the lives of multiple people at risk, certainty always wins out.

Tesla Hardware And Sneaky Radars

Arbe Phoenix radar module installed in a Tesla car as part of the Hardware 4 Autopilot hardware. (Credit: @greentheonly, Twitter)
One of the poorly kept secrets about Tesla’s Autopilot system is that it’s had a front-facing radar sensor for most of the time. Starting with Hardware 1 (HW1), it featured a single front-facing camera behind the top of the windshield and a radar behind the lower grille, in addition to 12 ultrasonic sensors around the vehicle.

Notable is that Tesla did not initially use the radar in a primary object detection role here, meaning that object detection and emergency stop functionality was performed using the RGB cameras. This changed after the RGB camera system failed to notice a white trailer against a bright sky, resulting in a spectacular crash. The subsequent firmware update gave the radar system the same role as the camera system, which likely would have prevented that particular crash.

HW1 used Mobileye’s EyeQ3, but after Mobileye cut ties with Tesla, NVidia’s Drive PX 2 was used instead for HW2. This upped the number of cameras to eight, providing a surround view of the car’s surroundings, with a similar forward-facing radar. After an intermedia HW2.5 revision, HW3 was the first to use a custom processor, featuring twelve Arm Cortex-A72 cores clocked at 2.6 GHz.

HW3 initially also had a radar sensor, but in 2021 this was eliminated with the ‘Tesla Vision’ system, which resulted in a significant uptick in crashes. In 2022 it was announced that the ultrasonic sensors for short-range object detection would be removed as well.

Then in January of 2023 HW4 started shipping, with even more impressive computing specs and 5 MP cameras instead of the previous 1.2 MP ones. This revision also reintroduced the forward-facing radar, apparently the Arbe Phoenix radar with a 300 meter range, but not in the Model Y. This indicates that RGB camera-only perception is still the primary mode for Tesla cars.

Answering The Question

At this point we can say with a high degree of certainty that by just using RGB cameras it is exceedingly hard to reliably stop a vehicle from smashing into objects, for the simple reason that you are reducing the amount of reliable data that goes into your decision-making software. While the object-detecting CNN may give a 29% possibility of an object being right up ahead, the radar or Lidar will have told you that a big, rather solid-looking object is lying on the road. Your own eyes would have told you that it’s a large piece of concrete that fell off a truck in front of you.

This then mostly leaves the question of whether the front-facing radar that’s present in at least some Tesla cars is about as good as the Lidar contraption that’s used by other car manufacturers like Volvo, as well as the roof-sized version by Waymo. After all, both work according to roughly the same basic principles.

That said, Lidar is superior when it comes to aspects like accuracy, as radar uses longer wavelengths. At the same time a radar system isn’t bothered as much by weather conditions, while generally being cheaper. For Waymo the choice for Lidar over radar comes down to this improved detail, as they can create a detailed 3D image of the surroundings, down to the direction that a pedestrian is facing, and hand signals by cyclists.

Thus the shortest possible answer is that yes, Lidar is absolutely the best option, while radar is a pretty good option to at least not drive into that semitrailer and/or pedestrian. Assuming your firmware is properly configured to act on said object detection, natch.

hackaday.com/2025/10/30/self-d…

100 pacchetti di Infostealer caricati su NPM sfruttando le allucinazioni delle AI

Da agosto 2024, la campagna PhantomRaven ha caricato 126 pacchetti dannosi su npm, che sono stati scaricati complessivamente oltre 86.000 volte. La campagna è stata scoperta da Koi Security, che ha riferito che gli attacchi sono stati abilitati da una funzionalità poco nota di npm che gli consente di aggirare la protezione e il rilevamento.

Si sottolinea che al momento della pubblicazione del rapporto erano ancora attivi circa 80 pacchetti dannosi. Gli esperti spiegano che gli aggressori sfruttano il meccanismo Remote Dynamic Dependencies (RDD).

In genere, uno sviluppatore vede tutte le dipendenze di un pacchetto in fase di installazione, scaricate dall’infrastruttura NPM attendibile. Tuttavia, RDD consente ai pacchetti di estrarre automaticamente il codice da URL esterni, anche tramite un canale HTTP non crittografato. Nel frattempo, il manifest del pacchetto non mostra alcuna dipendenza.

Quando uno sviluppatore esegue npm install, il pacchetto dannoso scarica silenziosamente un payload da un server controllato dagli aggressori e lo esegue immediatamente. Non è richiesta alcuna interazione da parte dell’utente e gli strumenti di analisi statica rimangono inconsapevoli dell’attività.

“PhantomRaven dimostra quanto possano essere sofisticati gli aggressori quando sfruttano i punti ciechi delle soluzioni di sicurezza tradizionali. Le dipendenze dinamiche remote sono semplicemente invisibili all’analisi statica”, affermano i ricercatori.

Si noti che il malware viene scaricato dal server ogni volta che il pacchetto viene installato, anziché essere memorizzato nella cache.

Questo apre le porte ad attacchi mirati: gli aggressori possono controllare l’indirizzo IP della richiesta e inviare codice innocuo ai ricercatori di sicurezza, distribuire codice dannoso per le reti aziendali e distribuire payload specializzati per gli ambienti cloud.

Una volta infettato, il malware raccoglie attentamente informazioni sul sistema della vittima:

• variabili di ambiente con configurazioni dei sistemi interni dello sviluppatore;
• token e credenziali per npm, GitHub Actions, GitLab, Jenkins e CircleCI;
• l’intero ambiente CI/CD attraverso il quale passano le modifiche al codice apportate da diversi sviluppatori.

I token rubati possono essere utilizzati per attaccare le supply chain e iniettare codice dannoso in progetti legittimi. Il furto di dati è organizzato in modo ridondante, utilizzando tre metodi: HTTP GET con dati nell’URL, HTTP POST con JSON e connessioni WebSocket.

Gli esperti scrivono che molti pacchetti dannosi sono mascherati da strumenti GitLab e Apache.

Lo slopsquatting, ovvero lo sfruttamento delle allucinazioni dell’intelligenza artificiale, gioca un ruolo speciale in questa campagna. Gli sviluppatori chiedono spesso agli assistenti LLM quali pacchetti siano più adatti a un particolare progetto. I modelli di intelligenza artificiale spesso inventano nomi inesistenti ma plausibili. Gli operatori PhantomRaven tracciano queste allucinazioni e registrano i pacchetti con questi nomi. Le vittime alla fine installano il malware da sole, seguendo le raccomandazioni di LLM.

Gli sviluppatori di LLM non comprendono ancora le cause esatte di queste allucinazioni e non sono in grado di creare modelli che le prevengano, ed è proprio questo che gli aggressori stanno sfruttando. I ricercatori ricordano di non affidarsi a LLM nella scelta delle dipendenze e di controllare attentamente i nomi dei pacchetti e le loro fonti, installando solo pacchetti provenienti da fornitori affidabili.

L'articolo 100 pacchetti di Infostealer caricati su NPM sfruttando le allucinazioni delle AI proviene da Red Hot Cyber.

Atroposia: la piattaforma MaaS che fornisce un Trojan munito di scanner delle vulnerabilità

I ricercatori di Varonis hanno scoperto la piattaforma MaaS (malware-as-a-service) Atroposia. Per 200 dollari al mese, i suoi clienti ricevono un Trojan di accesso remoto con funzionalità estese, tra cui desktop remoto, gestione del file system, furto di informazioni, credenziali, contenuto degli appunti, wallet di criptovalute, dirottamento DNS e uno scanner integrato per le vulnerabilità locali.

Secondo gli analisti, Atroposia ha un’architettura modulare. Il malware comunica con i server di comando e controllo tramite canali crittografati ed è in grado di bypassare il Controllo Account Utente (UAC) per aumentare i privilegi in Windows.

Una volta infettato, fornisce un accesso persistente e non rilevabile al sistema della vittima. I moduli chiave di Atroposia sono:

HRDP Connect avvia una sessione di desktop remoto nascosta in background, consentendo agli aggressori di aprire applicazioni, leggere documenti ed e-mail e, in generale, interagire con il sistema senza alcun segno visibile di attività dannosa. I ricercatori sottolineano che gli strumenti standard di monitoraggio dell’accesso remoto potrebbero “non rilevare” questa attività.

Il file manager funziona come un familiare Esplora risorse di Windows: gli aggressori possono visualizzare, copiare, eliminare ed eseguire i file. Il componente grabber cerca i dati per estensione o parola chiave, li comprime in archivi ZIP protetti da password e li invia al server di comando e controllo utilizzando metodi in-memory, riducendo al minimo le tracce dell’attacco sul sistema.

Stealer raccoglie dati di accesso salvati, dati del portafoglio di criptovalute e file di chat. Il gestore degli appunti intercetta tutto ciò che l’utente copia (password, chiavi API, indirizzi del portafoglio) in tempo reale e lo conserva per gli aggressori.

Il modulo di spoofing DNS sostituisce i domini con gli indirizzi IP degli aggressori a livello di host, reindirizzando silenziosamente le vittime verso server controllati dagli hacker. Questo apre le porte a phishing, attacchi MitM, falsi aggiornamenti, iniezione di adware o malware e furto di dati tramite query DNS.

Lo scanner di vulnerabilità integrato analizza il sistema della vittima alla ricerca di vulnerabilità non corrette, impostazioni non sicure e software obsoleto. I risultati vengono inviati agli operatori di malware sotto forma di punteggio, che gli aggressori possono utilizzare per pianificare ulteriori attacchi.

I ricercatori avvertono che questo modulo è particolarmente pericoloso negli ambienti aziendali: il malware potrebbe rilevare un client VPN obsoleto o una vulnerabilità di escalation dei privilegi, che può quindi essere sfruttata per ottenere informazioni più approfondite sull’infrastruttura della vittima. Inoltre, lo scanner analizza i sistemi vulnerabili nelle vicinanze per rilevare eventuali movimenti laterali.

Varonis osserva che Atroposia prosegue la tendenza verso la democratizzazione del crimine informatico.

Insieme ad altre piattaforme MaaS (come SpamGPT e MatrixPDF), riduce la barriera tecnica all’ingresso, consentendo anche ad aggressori poco qualificati di condurre efficaci “attacchi in abbonamento”.

L'articolo Atroposia: la piattaforma MaaS che fornisce un Trojan munito di scanner delle vulnerabilità proviene da Red Hot Cyber.