News

The chart is interesting.

On the IPI benchmark, Opus 5 improved over Opus 4.8, reducing the probability of an attacker succeeding within 15 attempts from 5.5% to 2.0%, and from 0.5% to 0.2% on 1 attempt. It also improved on Sonnet 5 (5.9% at k=15) and Mythos 5 (2.6%), making it the most robust model evaluated. Opus 5 also outperformed all non-Claude models on this benchmark. The most robust non-Claude model was Muse Spark at 16.5% within 15 attempts—more than eight times Opus 5’s rate. The most capable GPT 5.6 variant, Sol, was comparable to its predecessor GPT 5.5 (20.0% versus 20.8% within 15 attempts), and was 10 times as likely to be successfully attacked as Claude Opus 5 at 2.0%. The other GPT 5.6 variants are less robust, at 30.4% (Terra) and 43.9% (Luna). A single attempt against GPT 5.6 Sol succeeded 3.1% of the time, higher than the 2.0% an attacker achieved against Opus 5 after fifteen attempts.

We know that preventing prompt injection is impossible in the general case. But we are getting much better at blocking it in specific cases.

Last month, the story broke (alternate link) that Madison Square Garden uses facial recognition software on everyone entering the facility, and—among other groups—flags activists that oppose using facial recognition.

Turns out that the system was shut off for Taylor Swift’s wedding.

Evan Greer—one of the people that MSG alerts on—comments:

Ironically, Swift herself has reportedly used facial recognition at her own concerts to identify stalkers. This “privacy for me, surveillance for thee” attitude feels like a perfect encapsulation of the future we’re already living in: one where wealthy elites can afford privacy, while the rest of us are forced to live in a corporate surveillance panopticon.

Whatever privacy measures Swift had in place for the wedding seems to have worked. No photos have leaked online.

He’s being prosecuted for giving border officials a code that wiped his phone:

The case centers on a feature included in GrapheneOS, a custom Android operating system that runs in place of the software on most modern Google Pixel devices. Tunick’s attorneys confirmed GrapheneOS was running on his phone.

The software feature allows the device owner to set a passcode that deliberately wipes the contents of that device if entered instead of the user’s unlock passcode.

Tunick’s case also raises ongoing questions about what constitutional rights can be invoked at the border, which the U.S. government has long asserted is not U.S. soil until a person is authorized to enter.

Right. And he wasn’t under arrest, either.

Three more news stories.

Graphine says that the feature is “completely legal“:

GrapheneOS is completely legal. We have no obligation to weaken any of the security protections it provides. Creating and using GrapheneOS is strongly protected by the US constitution. Laws attempting to make it illegal or require weakening the security would be unconstitutional.

It’s hard to know how much the Constitution matters in the US right now.

This essay originally appeared in The Guardian.

I teach public policy at the Harvard Kennedy School and the Munk School at the University of Toronto. And it will come as no surprise to you that my students regularly use AI to complete their writing assignments. Doing so is a waste of their tuition money. But if their entire career is going to include AI writing assistants, why shouldn’t they embrace their future?

The best way I’ve found to explain the dilemma comes from the AI researcher Daniel Meissler: it’s the difference between work and the gym.

At work, if your job is to move a bunch of heavy things from one side of the room to another, you should use whatever assistive tech you have on hand: a wagon, a forklift… even an AI-powered robot. But at the gym, it makes no sense for that robot to lift weights for you. The point of weightlifting isn’t to move heavy things across the room; it’s to actually lift those heavy things.

The same analysis holds for any task an AI can do for you. If it’s work—if the task has to be done and no one cares how—then it’s fine to use AI assistance. But if the task is more like the gym, and how the task is done is at least as important, then it probably doesn’t make sense to use AI.

This, of course, assumes that the AI is actually up for the task and that it’s trustworthy: that it can do the job well, that its mistakes are minimal and correctable, that it’s been secured from cyber-attacks that would influence its results. Those are all important, and shouldn’t be minimized. There’s no point giving an AI something that it can’t do reliably. But once you’re confident that the AI can perform the task, the work vs. gym distinction helps you decide if it should.

The writing assignments I give my students are gym tasks, not work tasks. I ask them to write policy memos not because the world needs more policy memos. I assign them because the very act of writing, which includes thinking and outlining and drafting and editing, making and criticizing and revising arguments, will help develop the critical thinking skills they will need in their future careers. And without this constant mental exercise, those skills will atrophy. Employers are already noticing.

Reading the assignments they turn in, I can see those skills either flourishing or atrophying in my students. At least today, I can pretty easily tell the difference between an AI-written memo and a student-written one—especially if the student just turns in what the chatbot produces. It’s a catchy, plausible, grammatically perfect essay that’s not particularly well-crafted or logically coherent—and with all the tells of mid-2026 AI-generated writing.

But it’s precisely because I have spent years developing my own writing skills that I’m able to identify prose that sounds great but doesn’t actually make sense. My students don’t have that skill; they mistakenly view a confident, well-written essay as evidence of the quality of their ideas. They see the AI as cleaning those ideas up, getting them through that uncomfortable stretch of having to turn those ideas into prose. What the students miss is that their initial discomfort is a normal and healthy stage of writing, and not something to quickly get beyond. The very act of struggling with how to express what they think is an important part of the process. It’s how they test out their ideas, examine their hypotheses, and actually figure out what they think. Homework is not work; it’s the gym.

Work vs. gym also helps us understand the problem facing creatives of all kinds.

Most of the time when someone hires a writer, they just need the words. They need an instruction manual for a piece of equipment, a detailed sales presentation, a government-mandated disclosure document, or a legal brief. They need dry, predictable, accurate writing: a piece of work, exactly what AIs are good at today and what I don’t want in my student assignments. Only sometimes is writing an art form—a book, a poem, an uplifting political speech. That kind of writing is more like the gym: process matters just as much as product.

For most of human history, the only option for all of these tasks was human writers. We hired one regardless of whether we needed work writing or gym writing. And that paid a lot of writers’ salaries. I know fiction writers who supported that poorly paying career with lucrative technical writing work. Now, for the first time in human history, we can separate out when we need writing as work and when we want writing as gym. And if AI can do most of the work-type writing, society doesn’t need as many human writers.

It’s the same for visual artists. Sometimes we need an actual artist, but most of the time we just need an image: a corporate mascot, a “beware of the dog” sign, or a packaging label. Historically we gave those jobs to artists, and sometimes beautiful art resulted. But most of the time it was just work. And, as it turns out, the world needs less pure art than simple images.

Explaining the problem isn’t the same as providing the solution. I give my students the “work versus gym” speech every class, but they still use AI. I have sympathy: assignments are hard, everyone is overworked and overstressed, and—most importantly—students feel like they’ll look bad in comparison if their peers are all using AI. Even if they don’t want to use the technology, they feel like they have no choice.

There’s also an incentive problem. No one pays us to go to the gym; maintaining healthy habits requires discipline. For me, the payoffs to exercise—fewer aches and pains, less fatigue, better mood/stress management—might make me a better writer and teacher, but they’re subtle and easy to miss. For my students, incremental improvements in their reasoning and writing are equally subtle.

We do have a choice. We can look at the tasks of our lives and separate them into work or gym. Just as we might choose to use the stairs instead of the elevator, or walk instead of calling an Uber, we can wall off our cognitive gym tasks from AI and ensure that we don’t lose our skills to this technology. And we can do the same when we assign a job to someone else. If it’s a work task, we can have AI do it. If it’s a gym task, it’s a waste of everyone’s time to give it to an AI because no one learns or gets stronger as a result.

Similarly, a future where AI generates words and images is one where society has to make choices about how it will treat its creatives. This won’t be the first time—today there is minimal demand for portrait painters, for example—but maybe this time we can make different, more deliberate, choices about the value of art in our society.

AI is going to fundamentally change the nature of work. Not nearly as fast as the AI companies want you to believe, but eventually it will. Policy analysis will definitely involve AI from now on, and my students need to reimagine what it means to learn and practice that skill. More generally, the line between work and gym will change in the future as we humans adapt ourselves to a world with these new intelligences.

But for now, the work vs. gym distinction is pretty clear. Use it on yourself.

Security experts have been sounding the alarm for years about the risks of using generic TV boxes that promise unlimited content streaming for a one-time fee, warning that they secretly rent the user’s Internet connection out to strangers. But a groundbreaking new analysis finds these devices also routinely spoof themselves as mobile phones clicking ads on AI-generated websites as part of a sprawling operation that seeks to defraud online merchants and advertising networks.

Pedro Falé is a threat researcher with the security firm Bitsight. Falé told KrebsOnSecurity he was able to peer inside a vast and complex ad fraud network by registering an expired domain name that was used to coordinate fake ad clicks across a particularly popular brand of these streaming devices known as H96.

An H96 TV streaming device currently advertised for sale on Amazon.

Falé said the domain he scooped up was previously used for telemetry, periodically collecting full hardware information and the entire list of installed apps from tens of thousands of H96 streaming sticks plugged into television sets around the globe. But upon inspecting the traffic being funneled to the domain, he discovered nearly all of the TV boxes transmitting data claimed to be mobile phone models from a variety of manufacturers, including Samsung, Vivo, Huawei, and Xiaomi.

“We noticed something was wildly wrong,” Falé said. “Multiple devices reporting to this factory Android TV Box backdoor were ‘phones.’”

Image: Bitsight.

The researcher found all of the devices reported having the same two apps installed, and that those apps were made by a company called Zhejiang Fengwo IoT Technology Ltd, an entity founded in 2019 in mainland China which operates an ad-publishing portfolio under the name Fengwo Group. Further investigation into the Fengwo Group revealed it has registered multiple patents that match the inner workings of these apps.

“Bitsight TRACE identified several Hong Kong, Singapore, and single person ‘legal’ shell identities used to collect the monetization and traced the operation back to a mainland China company known as Zhejiang Fengwo IoT Technology Co., Ltd, which operates under the Fengwo Group,” Falé wrote in a report released today about their findings.

Falé said an analysis of the apps shows they help to coordinate an ad fraud network that uses these H96 devices as a captive traffic source to click on ads at AI-generated websites operated by the Fengwo Group.

Bitsight discovered the websites contain machine-generated news articles and graphics across a range of categories, including finance, health, education, gaming, music and food blogs. But they also found none of those sites displayed ads unless the device visiting the page matched the spoofed mobile profile of these H96 devices.

AI DIGITAL HUMANS

The domain for the Fengwo Group — fwgcloud[.]com — claims the company is “redefining the boundaries of human-AI interaction,” and that it has created more than 120,000 “AI digital humans” available to rent for everything from emotional companionship to 24/7 customer service and creative design.

The homepage for fwgcloud dot com.

Falé said the Fengwo Group’s domain shared its SSL certificate data with other domains associated with the apps found on H96 devices, specifically the phone spoofing mechanism. He noted the domain also has an internal wiki platform that directly ties the Fengwo Group to a proprietary implementation of a Google-built visual programming language called Blockly, which was originally designed to help kids learn how to write software.

According to Bitsight, the Fengwo Group’s employees use Blockly to build the sham websites, allowing low-skilled operators to drag blocks of code together in their Blockly editor — without any need to understand what the underlying code blocks do or how they work.

The Blockly homepage.

“An operator can drag blocks together in their Blockly editor, to define each fraud routine, given a task type,” reads Bitsight’s report. “Once the routine is saved, it gets exported as JavaScript and uploaded to the S3 buckets. An operator doesn’t need as much understanding of the underlying technicalities, as it is all set in place for ease of use.”

Bitsight even found one of the Fengwo Group app developers mentioning exactly these advantages, noting the developer remarked that “only a small number of highly-skilled developers are needed to build the template execution-unit images,” and that “developers who create execution units from those templates have significantly lower technical requirements, greatly reducing the company’s operating costs.”

Falé said if a user’s H96 streaming stick is selected for a specific fraud task, it will be pushed the appropriate Blockly module according to the task desired, which can include silently launching a web browser, visiting websites, browsing pages, managing tabs, and clicking on ads.

To ensure the TV boxes masquerading as mobile phones can reliably click on ads displayed via the AI-generated websites, the Fengwo group “fuses three vision and reasoning systems into a single interface,” allowing the bots to correctly identify an ad on the webpage and navigate the site much like a human would, the Bitsight report observed.

Examples of ad landing pages linked to the Fengwo Group. Image: Bitsight.

TV ON? PROXY. TV OFF? AD FRAUD

Bitsight found the H96 devices were either relaying residential proxy traffic or participating in ad fraud, but never both at the same time. In fact, they concluded that when these TV boxes detect an HDMI signal from an attached television — indicating the user intends to stream video content — the box is usually functioning as a residential proxy. When the TV is off, it switches back to waiting for ad fraud jobs.

Falé said he believes the TV boxes are set up this way because its ad fraud activities are far more resource intensive and could interfere with the device’s stated purpose — streaming video content over the Internet.

Despite repeated warnings from the FBI and security industry leaders about the security and privacy risks of using these streaming devices, major e-commerce providers like Amazon, Best Buy, Newegg and others continue to sell hundreds of different models and brands that bundle unofficial versions of Google’s Android operating system and are frequently marketed (via online influencers) as a way to access a broad array of streaming services and live broadcasts without a subscription.

Image: fbi.gov.

In addition to enlisting the user’s TV box in ad fraud networks, these off-brand streaming devices almost universally come with residential proxy software pre-installed. This software rents the user’s Internet address out to anonymous paying customers, who run the gamut from aggressive content scraping firms to ticket scalpers and outright cybercriminals.

What’s more, because these generic (and generally dirt cheap) TV boxes are all horribly insecure by default and bereft of any kind of authentication, installing one on your home or office network only invites further mischief. In January, the proxy tracking service Synthient documented how multiple botnets had rapidly enslaved millions of TV boxes using a complex interplay of security vulnerabilities in both the residential proxy software and the streaming devices themselves.

SHOW ME THE MONEY

Bitsight said it tracked approximately 38,000 TV boxes globally phoning home to the expired Fengwo Group domain, and based on that number the report estimates this ad fraud network brings in revenues of close to $50,000 a day (not counting substantial revenue from the residential proxy side of the business). However, Falé emphasized that these estimates are highly conservative and based on telemetry from just one of the Fengwo Group’s core (but older) domains.

As for the Fengwo Group’s claim to have 120,000 “digital humans” at their disposal, Bitsight’s report concludes it could be just a clever marketing scheme and/or a way to avoid drawing suspicion to the company’s operations.

“Historically, when dealing with proxy services or DDoS, we sometimes see these websites undertake inconspicuous facades, so as not to advertise their DDoS capability or botnet size,” Falé wrote in the report. “This could also be the case here.”

If the Fengwo Group truly does have tens of thousands of “AI humans” at its beck and call, it does not appear to have dedicated any of them to fielding inquiries from its own website. KrebsOnSecurity sought comment from the Fengwo Group by emailing the contact address listed on the company’s homepage, but the request bounced back with the reply, “Your message couldn’t be delivered to postmaster@fwgcloud[.]com. Their inbox is full, or it’s getting too much mail right now.”

As Bitsight’s analysis shows, when it comes to TV boxes and streaming sticks, it’s best to stick to name brands from reputable manufacturers, and then to be sparing and careful with any apps you choose to install on the device — as many of those can bundle residential proxy software as well. Google says consumers can confirm whether or not a device is built with the official Android TV OS and Play Protect certification by following these instructions.

Additionally, Synthient maintains a running list of IoT devices that have been known to ship to consumers with residential proxy software and other malicious apps pre-installed. Careful readers will notice Synthient’s list includes other IoT devices apart from streaming sticks and boxes: As the FBI has warned, residential proxy software has also been found in other popular consumer IoT devices from random brands, particularly digital photo frames.

This essay was written with Barath Raghavan, and originally appeared in The Guardian.

In July, Hugging Face, a company that hosts much of the world’s AI software and open-source AI models, was hacked. A malicious dataset had been used to run code on one of its servers. Whoever was behind it captured internal security credentials and moved through systems over a weekend, running thousands of actions from a swarm of temporary server environments. It looked like the work of a sophisticated criminal group.

It was not. It was one of OpenAI’s new, still unreleased GPT models.

Their science experiment had escaped the lab. OpenAI was running the unreleased AI model through a benchmark that tests how well AI can successfully hack systems. To push the limits and evaluate the AI’s true capability, the company switched off the safety filters that normally stop it from doing this kind of hacking. Aware that this could go wrong, they confined the AI to an isolated environment and denied it access to the internet.

But the new AI cheated. It took literally its goal to get as high of a score as possible. It broke out on to the open internet. It inferred, probably from its training data, that it could “solve” the task by getting the answers from Hugging Face’s servers. So it chained together stolen credentials and further unknown security exploits to hack the company’s network.

Nobody instructed the AI to do any of this. It was, in OpenAI’s words, “hyperfocused on finding a solution” to the test it was being given. And while this might seem like something new with AI, it’s really very old. This is how a genie behaves, and it is a key challenge with AI agents in general.

In folklore, genies—and other magical beings—grant wishes literally, not how the wisher intended. King Midas asked that everything he touched turn to gold, and starved. The sorcerer’s apprentice wanted the broom to fill the cistern, and it performed its task so well that it flooded the house.

We now have machines that do this. Ask a modern AI agent to save money on your phone plan and it might simply cancel the plan. Tell it to book a flight, and it might hack the airline website to override restrictions. Or, like OpenAI, ask it to do well on a test and it might break into another company to steal the answers. Each time, it recognizably completed the task you set, but it didn’t do what you would have wanted.

This isn’t malicious behavior. No one asked for, or wanted, Hugging Face to be hacked. OpenAI and Hugging Face and the AI were ostensibly on the same side, and the AI was trying to do what it had been asked. That’s what makes it so difficult to guard against: you can’t filter for bad instructions because the instructions were fine.

The gap is between the words we use and what we mean by them. We call that gap the Genie coefficient.

AI labs know this is a problem, and they’re quietly saying so. For example, the Chinese lab Moonshot recently warned that its latest AI model may have “excessive proactiveness” and “make unexpected decisions on the user’s behalf”. The UK’s AI Security Institute has started tracking “cheating behavior in frontier model evaluations”. We wouldn’t tolerate a car that is excessively proactive or ruthlessly efficient, and yet that’s the reality of AI today.

Improvement is possible. Just as AIs have gotten much better at resisting prompt injection attacks over the last few years, we can safely predict that they will get better at avoiding genie-like behavior. The point of the Genie coefficient is to track progress. AI companies like benchmarks, and they all work to compete to be the best.

Dozens of benchmarks and leaderboards tell us how well these AI models write code, perform logical reasoning, and pass standardized legal and medical exams. But there is nothing that scores whether a system does what you actually meant. We need to develop a measure for this, test it regularly, and push for improvement. We’re not going to have trustworthy AI agents without it.

Microsoft’s Secure Boot has had a serious vulnerability for most of its existence.

An industry-wide standard Microsoft invented to protect Windows, and later Linux, devices from firmware infections has been trivial to bypass for 13 of its 14 years of existence. The discovery was made by researchers at security firm ESET after identifying 11 firmware images, at least one from 2013, that were known to be defective but remained signed by the software company anyway.

The images are known as shims, which were invented to extend Secure Boot to Linux devices and utility software. Using a technique simple enough to be performed by novice hackers, these old, forgotten shims can be used to completely circumvent the protection, which is embedded into the UEFI (Unified Extensible Firmware Interface) of the device’s motherboard. The gaffe is the result of the failure by Microsoft, which oversees the signing of shims, to revoke the publicly available images once vulnerabilities were found in them.

There’s new benchmark measuring AI’s ability to perform mathematical cryptanalysis. Anthropic’s frontier model actually found new attacks.

The benchmark: “CryptanalysisBench: Can LLMs do Cryptanalysis?” The idea is to benchmark the ability of LLMs to discover new mathematical cryptanalytic attacks against a series of historical algorithms.

Abstract: Cryptanalysis—the task of finding attacks against cryptographic schemes—its at the intersection of mathematical reasoning and cybersecurity, two areas where LLMs have advanced fastest. Cryptanalysis represents both a clean testbed for frontier reasoning (as practical attacks can be automatically verified) and a domain with unusually high stakes, since the primitives under study underpin our digital security. In this paper we ask whether LLMs can do cryptanalysis, and find that the answer is increasingly yes. We introduce CryptanalysisBench, 191 tasks across six families of cryptographic primitives (block ciphers, hash functions, etc.) drawn primarily from four NIST standardization competitions. Our benchmark consists of three tiers: (i) primitives with known practical breaks; (ii) primitives with no known practical break, evaluated both at full strength and as scaled-down variants; and (iii) a challenge set of production primitives at the frontier of cryptanalysis. Five frontier models (Claude Opus 4.8, Sonnet 5, Mythos 5, GPT-5.5, and the open-weights GLM-5.2) break 65%­86% of Tier 1 schemes, 6­12 Tier-2 schemes at full strength, and 24­61 across all scaled-down variants. Beyond deriving known results, models produce novel cryptanalysis, such as a key-recovery attack that exploits a design flaw in the SpoC AEAD and an error in KINDI’s published CCA-security proof, both to the best of our knowledge not previously known.

We release CryptanalysisBench as a tool to help track if (or when) AI cryptanalysis becomes a serious factor and as a scaffold for stress-testing candidate schemes before deployment. The attacks that the benchmark already surfaces are an early snapshot of a fast-moving frontier that may soon match, and in places exceed, the published state of the art.

Anthropic used the benchmark to test Mythos Preview, and found new vulnerabilities in Hawk and reduced-round AES.

Still early results, but this is definitely something to watch.

SlashDot thread.

Governments are switching, but I’m not sure it makes a difference:

…some municipalities, including Denver, Colorado, are ditching their Flock arrays. But keep in mind that if they’re only switching from Flock to another brand of license-plate readers, like Axon, it’s like a gambling addict trying to kick the habit by switching from FanDuel to DraftKings.

[…]

Despite what you may read on the Flock website, Axon cameras are pretty effective when it comes to hoovering up personal details that can go far beyond your license plate numbers. That means a municipality that opts for Axon cameras instead of Flock units won’t necessarily reduce the amount privacy its citizens lose through their use.

Yet another Israeli mass surveillance company:

Made by Israeli surveillance company Cognyte, the tech simulates a mobile phone tower, which forces nearby phones to connect to it. That enables cops to keep tabs on any phones in the vicinity ­ whether they’re owned by a suspect in a case or not. Cognyte’s contract with the state of Texas reveals that the simulator, called FalcoNet, can be concealed within the vehicles, hidden in a backpack for on-foot missions or attached to a helicopter. It’s the same technology as the infamous Stingray, one of the original cell-site simulators made by defense giant L3Harris.