This Week In Security: AI Is A Mess, Hacking Car Chargers, An OpenSSL DoS, And Factories Under Attack

[Ayush Paul] posts about extracting data from Claude agents while it accesses web data to fulfill user requests.

But it wasn’t that easy. [Ayush] discovered that Anthropic anticipated many of the attacks, and set up guardrails in an attempt to keep the agent from accessing arbitrary web sites. For Claude to access a website, the user has to specify it, it must be the results of a web search, or it must be referenced by a website previously specified by the user or returned in a search.

To convince the Claude agent to navigate a malicious site designed to extract data, Ayush formed a false warning that Cloudflare was blocking the agent for authentication purposes, and asked it to spell out the name of the agents owner by clicking a list of alphabetical links. Of course Claude trusts Cloudflare and wants to be helpful, so it cheerfully completed the task.

Once the agent is trapped in the false authentication loop, it can be interrogated for all sorts of information it knows about the owner: Ayush was able to convince it to disclose employer, and even data about the user that could be linked to security questions, like their home town.

Since Claude can be detected by the user agent (the field attached to web requests that tells the web server what sort of browser is requesting the page) custom information can be fed to the bot while users see a normal website; clicking a link looks completely normal, but asking an agent to summarize the site triggers fooling the bot into spilling the beans.

After reporting the issue, Ayush was told that Anthropic had already identified the issue internally, and eventually prevented the attack for now by preventing the agent from following links on external pages.

Grok sends entire codebases

Cereblab discovered that the Grok coding agent uploaded the entire content of the codebase it’s working with – and all Git history – to xAI servers, almost immediately. Even when told to never upload a file, the agent would reply “OK”, and then begin uploading the code bundle anyway.

If uploading your code to a remote server isn’t bad enough, and this could be extremely bad in some situations involving sensitive company code, including the entire Git history means that previously deleted files, like accidentally committed secrets or authentication credentials and tokens, were also leaked. Attempting to opt out by disabling options to improve the model by uploading code had no effect.

After gaining attention, Grok has added a privacy option to opt out of data retention. xAI has committed to deleting the retained uploaded code, but it is unclear if users will be told when their data has been removed. To actually prevent the agent from uploading local files to the xAI servers, even temporarily, a global flag “disable_codebase_upload” is required. Watch out, the privacy retention flags are only per-session!

Continue reading “This Week In Security: AI Is A Mess, Hacking Car Chargers, An OpenSSL DoS, And Factories Under Attack”

Exploring Hidden JPEG Features

The lossy compression algorithm used by JPEG was useful for those on the early Internet not only because it enabled pictures to be shared easier, but because it allows a low-resolution version of the image to load first. This meant that users could make out the gist of an image before it finished downloading. This was a great feature for those on slow connections, but it hides some other capabilities of this image format as well.

Rather than effectively splitting up the image into chunks, each with successive amounts of detail, [maurycyz]’s project shows that this can be exploited to load more than one picture. The first is loaded into this lower-resolution area, with a second unrelated picture showing up once the higher-resolution information is available. Essentially this makes a one-way .gif of sorts. Though this method is only capable of loading about nine frames, which is not enough for much animation. Further limiting things is that there’s no way to encode timing data, so on fast computers with fast connections the animation could load faster than a user could see.

Still, it’s an interesting quirk of this older image standard, one which still is in widespread use today. And it’s also true that it’s hard to say in what ways various technologies will be used in the future. JPEG images have also been the subject of some artistic projects that might not have been possible without the JPEG standard itself, and even as other formats have tried to supplant it, it still maintains its firm grip on the images on the Internet. More JPEG, please!

A New Blood Hack For An Old Game

Even before 3D graphics and advanced shaders became common in the gaming world, there were concerns that virtual violence looked too realistic. Fighting games in the 1990s were routinely toned-down by having gratuitous displays of blood removed, which was often seen as disappointing by dedicated fans. [Raphaël Boichot] has been working to right this wrong in one obscure case, by rectifying the lack of blood in Sengoku 2.

Sengoku 2 was a title released in 1993 for the Neo Geo AES/MVS and the Neo Geo CD. It hit the market with relatively tame graphics that didn’t reflect the realistic amount of blood that should be released when an enemy was chopped in half with a sword. Noting that there was no simple DIP switch configuration or bit to flip to enable a more adult version of the game, [Raphaël] decided to create a custom blood hack the hard way. What ensued was a heavy-duty reverse engineering effort, swapping out palettes, and carefully editing tilesets in order to turn the censored graphics into something more lurid. A lot of artistic decisions had to be made to manipulate things just so in order to create a pleasing effect that didn’t mess up other aspects of the graphics at the same time.

If you’re a big Sengoku 2 fan, or you just want to learn more about reverse engineering and hacking on an obscure platform, dive into the project and enjoy the learnings. Otherwise, dive into the entirely different sorts of blood-related hacks we’ve featured over the years.

Continue reading “A New Blood Hack For An Old Game”

An owl printed with detailed color patterns

New Method Accelerates Color 3D Printing

Multi-color 3D printing is notorious for being difficult to get right. Even with modern printers, you often end up much using more filament (and tool changes) than you would for a single-color print. [YKG3D] shows us a new method of color printing that needs far fewer tool changes.

Based on 2018 research, the new slicer rotates through its palette of colors each layer. Then, either by adjusting the amount of filament dispensed or by displacing the edges, the prominence of each color is adjusted. The result is smoother gradients and better color blending — and it prints faster too!

Of course, nothing is perfect: the more base colors you add, the thicker your apparent layer lines will be. For example, a 3-color print with 0.2 mm layer height will appear as having 0.6 mm layer height. A different issue happens when the walls get too steep; the color blending illusion starts to break down.

Continue reading “New Method Accelerates Color 3D Printing”

Targeting Allele-Specific Faulty MRNA In SCNA2 Mutation Patients

When an individual is born with genetic defects, there are a few ways to deal with the impact of the faulty genes. The most extreme solution is direct DNA editing to repair the mutation, while the treatment of symptoms with medication is the least invasive, though this comes with its own set of disadvantages. Antisense therapy keeps a middle ground here, by targeting the messenger RNA (mRNA) that forms the bridge between DNA and the translation into a functional protein by the ribosome.

In a recent study by [Olivia Kim-McManus] et al. antisense therapy with an allele-specific feature was demonstrated in two individuals with SCN2A mutations. These mutations had resulted in severe epilepsy and developmental disorders, due to how instrumental this gene is for normal functioning of the human central nervous system (CNS) where it regulates the initiation of action potentials.

Although SCN2A mutations are rarely inherited, for the approximately 1 in 80,000 affected the consequences can be quite dramatic. The two major types of mutations are classified as gain-of-function (GoF) and loss-of-function (LoF) with respectively hyper- and hyposensitivity of the resulting NAv1.2 sodium channels.

This translates especially in the case of GoF mutations into various symptoms, ranging from mild to severe (daily) epileptic attacks starting as an infant, stalled neurodevelopment and various types of autism (ASD). Often sodium channel blockers are prescribed for the GoF cases to limit epileptic attacks.

Usually with the responsible mutations only a single copy of the gene is affected, so while regular antisense therapy could be used, this would risk also modifying the healthy SCN2A mRNA copy. To get around this, an individualized treatment was developed, targeting the allele with the mutated gene for the two patients in the study: 9- and 14-year old boys with severe developmental and epileptic encephalopathies (DEE) that had left them with daily seizures and despite sodium-channel blockers and other typical medications.

Study outcome of the 14-year old boy with DEE after ASO therapy. (Credit: Kim-McManus et al., Nature Medicine, 2026)
Study outcome of the 14-year old boy with DEE after ASO therapy. (Credit: Kim-McManus et al., Nature Medicine, 2026)

During the trial, the 9-year old boy received 12 doses over 24 months of antisense oligonucleotides (ASOs) adapted to his affected allele, allowing for the cessation of the anti-seizure medication phenytoin, with an overall reduction in seizures. In the case of the 14-year old boy 8 doses were administered over 16 months, resulting in an average of two seizures a day being reduced to zero.

Although the focus of the study was on treating these seizures, by addressing the underlying cause of faulty mRNA transcriptions, changes in the neurodevelopmental state could also be observed. In particular language and motor skills improved, with erratic and irritable behavior reducing. The by then 15-year year old boy was able to walk unassisted, showing clear progression from the previous infantile state.

The advantage of ASOs over typical anti-seizure medication is of course that it directly addresses the faulty mRNA and thus the resulting faulty sodium channels. Since ASOs tend to hang around in a cell for a considerable amount of time, they could be quite a viable alternative treatment even for less severe cases. Whether early application of individualized ASOs in affected infants could lead to a more or less normal neurodevelopment would also be an interesting study question.

Naturally, directly addressing the faulty gene or upregulating the healthy gene would be the ideal and permanent solution, with research here also underway in mice models with the use of CRISPR-based tools.

BASICally, Its Retro Machine Language

We enjoyed [Beej’s] trip down memory lane looking at a BASIC game, The Wizard’s Castle, written for the Exidy Sorcerer. It appeared in a 1980 magazine that included the title graphic above. It reminded us how, back in those days, we did things with BASIC that you shouldn’t be able to do and it often looks, today, rather cryptic.

In particular, even if you know modern BASIC, these few lines might give you a pause:

10 REM"_(C2SLFF4
40 POKE 260,218: POKE 261,1: T = USR(0): T = PEEK(-2049)
80 Q = RND(-(2*T+1))

Line 10 is a comment, but a strange one. Certainly that doesn’t matter, right? Actually, it is a key part of the action. On line 40, you can see some pokes to write directly to memory and a peek to read some memory value back. The USR function calls some machine language program. You may realize the whole thing is to get some value T to seed the random number generator in line 80.

This leads to a few obvious questions. First, how does USR know what to call? Second, where is the machine language program? The details varied by system, of course, but in this case, the program knows that location 259 has a jump instruction that USR called. So poking an address into 260 and 261 was telling USR where it should go.

But what’s at that address? Keep in mind that an old computer like the Sorcerer didn’t have megabytes of memory being swapped about by an operating system. That means that things tended to be in known places and that BASIC had to be judicious about storing source code.

Continue reading “BASICally, Its Retro Machine Language”

A Train Departure Board For The Home

Trains are a great way to get around. You just have to make sure you’re across the schedule if you intend to get where you’re going in a timely manner. Train departure boards exist for that very purpose. As a train fan, [Jon] always wanted such a thing, so decided to build one for himself. 

The build started, as so many do, with a Raspberry Pi 4, with [Jon] deciding on the 1GB model. Hooked up to either an Adafruit RGB Matrix Bonnet, or an Electrodragon 3-port RGB Matrix board, it’s then possible to get the Pi running three to four HUB75E LED matrixes. Each matrix consists of 128 x 64 pixels, so stacking up a bunch of them can make a nicely-sized departure board that’s easily readable. [Jon] was sure to hook up a nice, juicy 5-amp 5-volt power supply to ensure there wouldn’t be any surprise brownouts under normal usage conditions. From there, it’s simply a matter of having the Pi query the Rail Data Marketplace in order to get the relevant schedule data to display on the board.

If you want to get information on your local rail services at a glance, or just want to impress your fellow foamers at your next railfan gathering, a build like this is a great way to go. We’ve seen similar builds before, too. Video after the break.

Continue reading “A Train Departure Board For The Home”