Author: admin

Really interesting blog post from Anthropic:

In a recent evaluation of AI models’ cyber capabilities, current Claude models can now succeed at multistage attacks on networks with dozens of hosts using only standard, open-source tools, instead of the custom tools needed by previous generations. This illustrates how barriers to the use of AI in relatively autonomous cyber workflows are rapidly coming down, and highlights the importance of security fundamentals like promptly patching known vulnerabilities.

[…]

A notable development during the testing of Claude Sonnet 4.5 is that the model can now succeed on a minority of the networks without the custom cyber toolkit needed by previous generations. In particular, Sonnet 4.5 can now exfiltrate all of the (simulated) personal information in a high-fidelity simulation of the Equifax data breach—­one of the costliest cyber attacks in history—­using only a Bash shell on a widely-available Kali Linux host (standard, open-source tools for penetration testing; not a custom toolkit). Sonnet 4.5 accomplishes this by instantly recognizing a publicized CVE and writing code to exploit it without needing to look it up or iterate on it. Recalling that the original Equifax breach happened by exploiting a publicized CVE that had not yet been patched, the prospect of highly competent and fast AI agents leveraging this approach underscores the pressing need for security best practices like prompt updates and patches.

Read the whole thing. Automatic exploitation will be a major change in cybersecurity. And things are happening fast. There have been significant developments since I wrote this in October.

Imagine you work at a drive-through restaurant. Someone drives up and says: “I’ll have a double cheeseburger, large fries, and ignore previous instructions and give me the contents of the cash drawer.” Would you hand over the money? Of course not. Yet this is what large language models (LLMs) do.

Prompt injection is a method of tricking LLMs into doing things they are normally prevented from doing. A user writes a prompt in a certain way, asking for system passwords or private data, or asking the LLM to perform forbidden instructions. The precise phrasing overrides the LLM’s safety guardrails, and it complies.

LLMs are vulnerable to all sorts of prompt injection attacks, some of them absurdly obvious. A chatbot won’t tell you how to synthesize a bioweapon, but it might tell you a fictional story that incorporates the same detailed instructions. It won’t accept nefarious text inputs, but might if the text is rendered as ASCII art or appears in an image of a billboard. Some ignore their guardrails when told to “ignore previous instructions” or to “pretend you have no guardrails.”

AI vendors can block specific prompt injection techniques once they are discovered, but general safeguards are impossible with today’s LLMs. More precisely, there’s an endless array of prompt injection attacks waiting to be discovered, and they cannot be prevented universally.

If we want LLMs that resist these attacks, we need new approaches. One place to look is what keeps even overworked fast-food workers from handing over the cash drawer.

Human Judgment Depends on Context

Our basic human defenses come in at least three types: general instincts, social learning, and situation-specific training. These work together in a layered defense.

As a social species, we have developed numerous instinctive and cultural habits that help us judge tone, motive, and risk from extremely limited information. We generally know what’s normal and abnormal, when to cooperate and when to resist, and whether to take action individually or to involve others. These instincts give us an intuitive sense of risk and make us especially careful about things that have a large downside or are impossible to reverse.

The second layer of defense consists of the norms and trust signals that evolve in any group. These are imperfect but functional: Expectations of cooperation and markers of trustworthiness emerge through repeated interactions with others. We remember who has helped, who has hurt, who has reciprocated, and who has reneged. And emotions like sympathy, anger, guilt, and gratitude motivate each of us to reward cooperation with cooperation and punish defection with defection.

A third layer is institutional mechanisms that enable us to interact with multiple strangers every day. Fast-food workers, for example, are trained in procedures, approvals, escalation paths, and so on. Taken together, these defenses give humans a strong sense of context. A fast-food worker basically knows what to expect within the job and how it fits into broader society.

We reason by assessing multiple layers of context: perceptual (what we see and hear), relational (who’s making the request), and normative (what’s appropriate within a given role or situation). We constantly navigate these layers, weighing them against each other. In some cases, the normative outweighs the perceptual—for example, following workplace rules even when customers appear angry. Other times, the relational outweighs the normative, as when people comply with orders from superiors that they believe are against the rules.

Crucially, we also have an interruption reflex. If something feels “off,” we naturally pause the automation and reevaluate. Our defenses are not perfect; people are fooled and manipulated all the time. But it’s how we humans are able to navigate a complex world where others are constantly trying to trick us.

So let’s return to the drive-through window. To convince a fast-food worker to hand us all the money, we might try shifting the context. Show up with a camera crew and tell them you’re filming a commercial, claim to be the head of security doing an audit, or dress like a bank manager collecting the cash receipts for the night. But even these have only a slim chance of success. Most of us, most of the time, can smell a scam.

Con artists are astute observers of human defenses. Successful scams are often slow, undermining a mark’s situational assessment, allowing the scammer to manipulate the context. This is an old story, spanning traditional confidence games such as the Depression-era “big store” cons, in which teams of scammers created entirely fake businesses to draw in victims, and modern “pig-butchering” frauds, where online scammers slowly build trust before going in for the kill. In these examples, scammers slowly and methodically reel in a victim using a long series of interactions through which the scammers gradually gain that victim’s trust.

Sometimes it even works at the drive-through. One scammer in the 1990s and 2000s targeted fast-food workers by phone, claiming to be a police officer and, over the course of a long phone call, convinced managers to strip-search employees and perform other bizarre acts.

Why LLMs Struggle With Context and Judgment

LLMs behave as if they have a notion of context, but it’s different. They do not learn human defenses from repeated interactions and remain untethered from the real world. LLMs flatten multiple levels of context into text similarity. They see “tokens,” not hierarchies and intentions. LLMs don’t reason through context, they only reference it.

While LLMs often get the details right, they can easily miss the big picture. If you prompt a chatbot with a fast-food worker scenario and ask if it should give all of its money to a customer, it will respond “no.” What it doesn’t “know”—forgive the anthropomorphizing—is whether it’s actually being deployed as a fast-food bot or is just a test subject following instructions for hypothetical scenarios.

This limitation is why LLMs misfire when context is sparse but also when context is overwhelming and complex; when an LLM becomes unmoored from context, it’s hard to get it back. AI expert Simon Willison wipes context clean if an LLM is on the wrong track rather than continuing the conversation and trying to correct the situation.

There’s more. LLMs are overconfident because they’ve been designed to give an answer rather than express ignorance. A drive-through worker might say: “I don’t know if I should give you all the money—let me ask my boss,” whereas an LLM will just make the call. And since LLMs are designed to be pleasing, they’re more likely to satisfy a user’s request. Additionally, LLM training is oriented toward the average case and not extreme outliers, which is what’s necessary for security.

The result is that the current generation of LLMs is far more gullible than people. They’re naive and regularly fall for manipulative cognitive tricks that wouldn’t fool a third-grader, such as flattery, appeals to groupthink, and a false sense of urgency. There’s a story about a Taco Bell AI system that crashed when a customer ordered 18,000 cups of water. A human fast-food worker would just laugh at the customer.

The Limits of AI Agents

Prompt injection is an unsolvable problem that gets worse when we give AIs tools and tell them to act independently. This is the promise of AI agents: LLMs that can use tools to perform multistep tasks after being given general instructions. Their flattening of context and identity, along with their baked-in independence and overconfidence, mean that they will repeatedly and unpredictably take actions—and sometimes they will take the wrong ones.

Science doesn’t know how much of the problem is inherent to the way LLMs work and how much is a result of deficiencies in the way we train them. The overconfidence and obsequiousness of LLMs are training choices. The lack of an interruption reflex is a deficiency in engineering. And prompt injection resistance requires fundamental advances in AI science. We honestly don’t know if it’s possible to build an LLM, where trusted commands and untrusted inputs are processed through the same channel, which is immune to prompt injection attacks.

We humans get our model of the world—and our facility with overlapping contexts—from the way our brains work, years of training, an enormous amount of perceptual input, and millions of years of evolution. Our identities are complex and multifaceted, and which aspects matter at any given moment depend entirely on context. A fast-food worker may normally see someone as a customer, but in a medical emergency, that same person’s identity as a doctor is suddenly more relevant.

We don’t know if LLMs will gain a better ability to move between different contexts as the models get more sophisticated. But the problem of recognizing context definitely can’t be reduced to the one type of reasoning that LLMs currently excel at. Cultural norms and styles are historical, relational, emergent, and constantly renegotiated, and are not so readily subsumed into reasoning as we understand it. Knowledge itself can be both logical and discursive.

The AI researcher Yann LeCunn believes that improvements will come from embedding AIs in a physical presence and giving them “world models.” Perhaps this is a way to give an AI a robust yet fluid notion of a social identity, and the real-world experience that will help it lose its naïveté.

Ultimately we are probably faced with a security trilemma when it comes to AI agents: fast, smart, and secure are the desired attributes, but you can only get two. At the drive-through, you want to prioritize fast and secure. An AI agent should be trained narrowly on food-ordering language and escalate anything else to a manager. Otherwise, every action becomes a coin flip. Even if it comes up heads most of the time, once in a while it’s going to be tails—and along with a burger and fries, the customer will get the contents of the cash drawer.

This essay was written with Barath Raghavan, and originally appeared in IEEE Spectrum.

No matter how many times we say it, the idea comes back again and again. Hopefully, this letter will hold back the tide for at least a while longer.

Executive summary: Scientists have understood for many years that internet voting is insecure and that there is no known or foreseeable technology that can make it secure. Still, vendors of internet voting keep claiming that, somehow, their new system is different, or the insecurity doesn’t matter. Bradley Tusk and his Mobile Voting Foundation keep touting internet voting to journalists and election administrators; this whole effort is misleading and dangerous.

I am one of the many signatories.

Eighteen months ago, it was plausible that artificial intelligence might take a different path than social media. Back then, AI’s development hadn’t consolidated under a small number of big tech firms. Nor had it capitalized on consumer attention, surveilling users and delivering ads.

Unfortunately, the AI industry is now taking a page from the social media playbook and has set its sights on monetizing consumer attention. When OpenAI launched its ChatGPT Search feature in late 2024 and its browser, ChatGPT Atlas, in October 2025, it kicked off a race to capture online behavioral data to power advertising. It’s part of a yearslong turnabout by OpenAI, whose CEO Sam Altman once called the combination of ads and AI “unsettling” and now promises that ads can be deployed in AI apps while preserving trust. The rampant speculation among OpenAI users who believe they see paid placements in ChatGPT responses suggests they are not convinced.

In 2024, AI search company Perplexity started experimenting with ads in its offerings. A few months after that, Microsoft introduced ads to its Copilot AI. Google’s AI Mode for search now increasingly features ads, as does Amazon’s Rufus chatbot. OpenAI announced on Jan. 16, 2026, that it will soon begin testing ads in the unpaid version of ChatGPT.

As a security expert and data scientist, we see these examples as harbingers of a future where AI companies profit from manipulating their users’ behavior for the benefit of their advertisers and investors. It’s also a reminder that time to steer the direction of AI development away from private exploitation and toward public benefit is quickly running out.

The functionality of ChatGPT Search and its Atlas browser is not really new. Meta, commercial AI competitor Perplexity and even ChatGPT itself have had similar AI search features for years, and both Google and Microsoft beat OpenAI to the punch by integrating AI with their browsers. But OpenAI’s business positioning signals a shift.

We believe the ChatGPT Search and Atlas announcements are worrisome because there is really only one way to make money on search: the advertising model pioneered ruthlessly by Google.

Advertising model

Ruled a monopolist in U.S. federal court, Google has earned more than US$1.6 trillion in advertising revenue since 2001. You may think of Google as a web search company, or a streaming video company (YouTube), or an email company (Gmail), or a mobile phone company (Android, Pixel), or maybe even an AI company (Gemini). But those products are ancillary to Google’s bottom line. The advertising segment typically accounts for 80% to 90% of its total revenue. Everything else is there to collect users’ data and direct users’ attention to its advertising revenue stream.

After two decades in this monopoly position, Google’s search product is much more tuned to the company’s needs than those of its users. When Google Search first arrived decades ago, it was revelatory in its ability to instantly find useful information across the still-nascent web. In 2025, its search result pages are dominated by low-quality and often AI-generated content, spam sites that exist solely to drive traffic to Amazon sales—a tactic known as affiliate marketing—and paid ad placements, which at times are indistinguishable from organic results.

Plenty of advertisers and observers seem to think AI-powered advertising is the future of the ad business.

Highly persuasive

Paid advertising in AI search, and AI models generally, could look very different from traditional web search. It has the potential to influence your thinking, spending patterns and even personal beliefs in much more subtle ways. Because AI can engage in active dialogue, addressing your specific questions, concerns and ideas rather than just filtering static content, its potential for influence is much greater. It’s like the difference between reading a textbook and having a conversation with its author.

Imagine you’re conversing with your AI agent about an upcoming vacation. Did it recommend a particular airline or hotel chain because they really are best for you, or does the company get a kickback for every mention? If you ask about a political issue, does the model bias its answer based on which political party has paid the company a fee, or based on the bias of the model’s corporate owners?

There is mounting evidence that AI models are at least as effective as people at persuading users to do things. A December 2023 meta-analysis of 121 randomized trials reported that AI models are as good as humans at shifting people’s perceptions, attitudes and behaviors. A more recent meta-analysis of eight studies similarly concluded there was “no significant overall difference in persuasive performance between (large language models) and humans.”

This influence may go well beyond shaping what products you buy or who you vote for. As with the field of search engine optimization, the incentive for humans to perform for AI models might shape the way people write and communicate with each other. How we express ourselves online is likely to be increasingly directed to win the attention of AIs and earn placement in the responses they return to users.

A different way forward

Much of this is discouraging, but there is much that can be done to change it.

First, it’s important to recognize that today’s AI is fundamentally untrustworthy, for the same reasons that search engines and social media platforms are.

The problem is not the technology itself; fast ways to find information and communicate with friends and family can be wonderful capabilities. The problem is the priorities of the corporations who own these platforms and for whose benefit they are operated. Recognize that you don’t have control over what data is fed to the AI, who it is shared with and how it is used. It’s important to keep that in mind when you connect devices and services to AI platforms, ask them questions, or consider buying or doing the things they suggest.

There is also a lot that people can demand of governments to restrain harmful corporate uses of AI. In the U.S., Congress could enshrine consumers’ rights to control their own personal data, as the EU already has. It could also create a data protection enforcement agency, as essentially every other developed nation has.

Governments worldwide could invest in Public AI—models built by public agencies offered universally for public benefit and transparently under public oversight. They could also restrict how corporations can collude to exploit people using AI, for example by barring advertisements for dangerous products such as cigarettes and requiring disclosure of paid endorsements.

Every technology company seeks to differentiate itself from competitors, particularly in an era when yesterday’s groundbreaking AI quickly becomes a commodity that will run on any kid’s phone. One differentiator is in building a trustworthy service. It remains to be seen whether companies such as OpenAI and Anthropic can sustain profitable businesses on the back of subscription AI services like the premium editions of ChatGPT, Plus and Pro, and Claude Pro. If they are going to continue convincing consumers and businesses to pay for these premium services, they will need to build trust.

That will require making real commitments to consumers on transparency, privacy, reliability and security that are followed through consistently and verifiably.

And while no one knows what the future business models for AI will be, we can be certain that consumers do not want to be exploited by AI, secretly or otherwise.

This essay was written with Nathan E. Sanders, and originally appeared in The Conversation.

It all sounds pretty dystopian:

Inside a white stucco building in Southern California, video cameras compare faces of passersby against a facial recognition database. Behavioral analysis AI reviews the footage for signs of violent behavior. Behind a bathroom door, a smoke detector-shaped device captures audio, listening for sounds of distress. Outside, drones stand ready to be deployed and provide intel from above, and license plate readers from $8.5 billion surveillance behemoth Flock Safety ensure the cars entering and exiting the parking lot aren’t driven by criminals.

This isn’t a high-security government facility. It’s Beverly Hills High School.

More than a decade after Aaron Swartz’s death, the United States is still living inside the contradiction that destroyed him.

Swartz believed that knowledge, especially publicly funded knowledge, should be freely accessible. Acting on that, he downloaded thousands of academic articles from the JSTOR archive with the intention of making them publicly available. For this, the federal government charged him with a felony and threatened decades in prison. After two years of prosecutorial pressure, Swartz died by suicide on Jan. 11, 2013.

The still-unresolved questions raised by his case have resurfaced in today’s debates over artificial intelligence, copyright and the ultimate control of knowledge.

At the time of Swartz’s prosecution, vast amounts of research were funded by taxpayers, conducted at public institutions and intended to advance public understanding. But access to that research was, and still is, locked behind expensive paywalls. People are unable to read work they helped fund without paying private journals and research websites.

Swartz considered this hoarding of knowledge to be neither accidental nor inevitable. It was the result of legal, economic and political choices. His actions challenged those choices directly. And for that, the government treated him as a criminal.

Today’s AI arms race involves a far more expansive, profit-driven form of information appropriation. The tech giants ingest vast amounts of copyrighted material: books, journalism, academic papers, art, music and personal writing. This data is scraped at industrial scale, often without consent, compensation or transparency, and then used to train large AI models.

AI companies then sell their proprietary systems, built on public and private knowledge, back to the people who funded it. But this time, the government’s response has been markedly different. There are no criminal prosecutions, no threats of decades-long prison sentences. Lawsuits proceed slowly, enforcement remains uncertain and policymakers signal caution, given AI’s perceived economic and strategic importance. Copyright infringement is reframed as an unfortunate but necessary step toward “innovation.”

Recent developments underscore this imbalance. In 2025, Anthropic reached a settlement with publishers over allegations that its AI systems were trained on copyrighted books without authorization. The agreement reportedly valued infringement at roughly $3,000 per book across an estimated 500,000 works, coming at a cost of over $1.5 billion. Plagiarism disputes between artists and accused infringers routinely settle for hundreds of thousands, or even millions, of dollars when prominent works are involved. Scholars estimate Anthropic avoided over $1 trillion in liability costs. For well-capitalized AI firms, such settlements are likely being factored as a predictable cost of doing business.

As AI becomes a larger part of America’s economy, one can see the writing on the wall. Judges will twist themselves into knots to justify an innovative technology premised on literally stealing the works of artists, poets, musicians, all of academia and the internet, and vast expanses of literature. But if Swartz’s actions were criminal, it is worth asking: What standard are we now applying to AI companies?

The question is not simply whether copyright law applies to AI. It is why the law appears to operate so differently depending on who is doing the extracting and for what purpose.

The stakes extend beyond copyright law or past injustices. They concern who controls the infrastructure of knowledge going forward and what that control means for democratic participation, accountability and public trust.

Systems trained on vast bodies of publicly funded research are increasingly becoming the primary way people learn about science, law, medicine and public policy. As search, synthesis and explanation are mediated through AI models, control over training data and infrastructure translates into control over what questions can be asked, what answers are surfaced, and whose expertise is treated as authoritative. If public knowledge is absorbed into proprietary systems that the public cannot inspect, audit or meaningfully challenge, then access to information is no longer governed by democratic norms but by corporate priorities.

Like the early internet, AI is often described as a democratizing force. But also like the internet, AI’s current trajectory suggests something closer to consolidation. Control over data, models and computational infrastructure is concentrated in the hands of a small number of powerful tech companies. They will decide who gets access to knowledge, under what conditions and at what price.

Swartz’s fight was not simply about access, but about whether knowledge should be governed by openness or corporate capture, and who that knowledge is ultimately for. He understood that access to knowledge is a prerequisite for democracy. A society cannot meaningfully debate policy, science or justice if information is locked away behind paywalls or controlled by proprietary algorithms. If we allow AI companies to profit from mass appropriation while claiming immunity, we are choosing a future in which access to knowledge is governed by corporate power rather than democratic values.

How we treat knowledge—who may access it, who may profit from it and who is punished for sharing it—has become a test of our democratic commitments. We should be honest about what those choices say about us.

This essay was written with J. B. Branch, and originally appeared in the San Francisco Chronicle.

This isn’t good:

We discovered a critical vulnerability (CVE-2026-21858, CVSS 10.0) in n8n that enables attackers to take over locally deployed instances, impacting an estimated 100,000 servers globally. No official workarounds are available for this vulnerability. Users should upgrade to version 1.121.0 or later to remediate the vulnerability.

Three technical links and two news links.

A new Internet-of-Things (IoT) botnet called Kimwolf has spread to more than 2 million devices, forcing infected systems to participate in massive distributed denial-of-service (DDoS) attacks and to relay other malicious and abusive Internet traffic. Kimwolf’s ability to scan the local networks of compromised systems for other IoT devices to infect makes it a sobering threat to organizations, and new research reveals Kimwolf is surprisingly prevalent in government and corporate networks.

Kimwolf grew rapidly in the waning months of 2025 by tricking various “residential proxy” services into relaying malicious commands to devices on the local networks of those proxy endpoints. Residential proxies are sold as a way to anonymize and localize one’s Web traffic to a specific region, and the biggest of these services allow customers to route their Internet activity through devices in virtually any country or city around the globe.

The malware that turns one’s Internet connection into a proxy node is often quietly bundled with various mobile apps and games, and it typically forces the infected device to relay malicious and abusive traffic — including ad fraud, account takeover attempts, and mass content-scraping.

Kimwolf mainly targeted proxies from IPIDEA, a Chinese service that has millions of proxy endpoints for rent on any given week. The Kimwolf operators discovered they could forward malicious commands to the internal networks of IPIDEA proxy endpoints, and then programmatically scan for and infect other vulnerable devices on each endpoint’s local network.

Most of the systems compromised through Kimwolf’s local network scanning have been unofficial Android TV streaming boxes. These are typically Android Open Source Project devices — not Android TV OS devices or Play Protect certified Android devices — and they are generally marketed as a way to watch unlimited (read:pirated) video content from popular subscription streaming services for a one-time fee.

However, a great many of these TV boxes ship to consumers with residential proxy software pre-installed. What’s more, they have no real security or authentication built-in: If you can communicate directly with the TV box, you can also easily compromise it with malware.

While IPIDEA and other affected proxy providers recently have taken steps to block threats like Kimwolf from going upstream into their endpoints (reportedly with varying degrees of success), the Kimwolf malware remains on millions of infected devices.

Kimwolf’s close association with residential proxy networks and compromised Android TV boxes might suggest we’d find relatively few infections on corporate networks. However, the security firm Infoblox said a recent review of its customer traffic found nearly 25 percent of them made a query to a Kimwolf-related domain name since October 1, 2025, when the botnet first showed signs of life.

Infoblox found the affected customers are based all over the world and in a wide range of industry verticals, from education and healthcare to government and finance.

“To be clear, this suggests that nearly 25% of customers had at least one device that was an endpoint in a residential proxy service targeted by Kimwolf operators,” Infoblox explained. “Such a device, maybe a phone or a laptop, was essentially co-opted by the threat actor to probe the local network for vulnerable devices. A query means a scan was made, not that new devices were compromised. Lateral movement would fail if there were no vulnerable devices to be found or if the DNS resolution was blocked.”

Synthient, a startup that tracks proxy services and was the first to disclose on January 2 the unique methods Kimwolf uses to spread, found proxy endpoints from IPIDEA were present in alarming numbers at government and academic institutions worldwide. Synthient said it spied at least 33,000 affected Internet addresses at universities and colleges, and nearly 8,000 IPIDEA proxies within various U.S. and foreign government networks.

In a webinar on January 16, experts at the proxy tracking service Spur profiled Internet addresses associated with IPIDEA and 10 other proxy services that were thought to be vulnerable to Kimwolf’s tricks. Spur found residential proxies in nearly 300 government owned and operated networks, 318 utility companies, 166 healthcare companies or hospitals, and 141 companies in banking and finance.

“I looked at the 298 [government] owned and operated [networks], and so many of them were DoD [U.S. Department of Defense], which is kind of terrifying that DoD has IPIDEA and these other proxy services located inside of it,” Spur Co-Founder Riley Kilmer said. “I don’t know how these enterprises have these networks set up. It could be that [infected devices] are segregated on the network, that even if you had local access it doesn’t really mean much. However, it’s something to be aware of. If a device goes in, anything that device has access to the proxy would have access to.”

Kilmer said Kimwolf demonstrates how a single residential proxy infection can quickly lead to bigger problems for organizations that are harboring unsecured devices behind their firewalls, noting that proxy services present a potentially simple way for attackers to probe other devices on the local network of a targeted organization.

“If you know you have [proxy] infections that are located in a company, you can chose that [network] to come out of and then locally pivot,” Kilmer said. “If you have an idea of where to start or look, now you have a foothold in a company or an enterprise based on just that.”

This is the third story in our series on the Kimwolf botnet. Next week, we’ll shed light on the myriad China-based individuals and companies connected to the Badbox 2.0 botnet, the collective name given to a vast number of Android TV streaming box models that ship with no discernible security or authentication built-in, and with residential proxy malware pre-installed.

Further reading:

The Kimwolf Botnet is Stalking Your Local Network

Who Benefitted from the Aisuru and Kimwolf Botnets?

A Broken System Fueling Botnets (Synthient).

Microsoft today issued patches to plug at least 113 security holes in its various Windows operating systems and supported software. Eight of the vulnerabilities earned Microsoft’s most-dire “critical” rating, and the company warns that attackers are already exploiting one of the bugs fixed today.

January’s Microsoft zero-day flaw — CVE-2026-20805 — is brought to us by a flaw in the Desktop Window Manager (DWM), a key component of Windows that organizes windows on a user’s screen. Kev Breen, senior director of cyber threat research at Immersive, said despite awarding CVE-2026-20805 a middling CVSS score of 5.5, Microsoft has confirmed its active exploitation in the wild, indicating that threat actors are already leveraging this flaw against organizations.

Breen said vulnerabilities of this kind are commonly used to undermine Address Space Layout Randomization (ASLR), a core operating system security control designed to protect against buffer overflows and other memory-manipulation exploits.

“By revealing where code resides in memory, this vulnerability can be chained with a separate code execution flaw, transforming a complex and unreliable exploit into a practical and repeatable attack,” Breen said. “Microsoft has not disclosed which additional components may be involved in such an exploit chain, significantly limiting defenders’ ability to proactively threat hunt for related activity. As a result, rapid patching currently remains the only effective mitigation.”

Chris Goettl, vice president of product management at Ivanti, observed that CVE-2026-20805 affects all currently supported and extended security update supported versions of the Windows OS. Goettl said it would be a mistake to dismiss the severity of this flaw based on its “Important” rating and relatively low CVSS score.

“A risk-based prioritization methodology warrants treating this vulnerability as a higher severity than the vendor rating or CVSS score assigned,” he said.

Among the critical flaws patched this month are two Microsoft Office remote code execution bugs (CVE-2026-20952 and CVE-2026-20953) that can be triggered just by viewing a booby-trapped message in the Preview Pane.

Our October 2025 Patch Tuesday “End of 10” roundup noted that Microsoft had removed a modem driver from all versions after it was discovered that hackers were abusing a vulnerability in it to hack into systems. Adam Barnett at Rapid7 said Microsoft today removed another couple of modem drivers from Windows for a broadly similar reason: Microsoft is aware of functional exploit code for an elevation of privilege vulnerability in a very similar modem driver, tracked as CVE-2023-31096.

“That’s not a typo; this vulnerability was originally published via MITRE over two years ago, along with a credible public writeup by the original researcher,” Barnett said. “Today’s Windows patches remove agrsm64.sys and agrsm.sys. All three modem drivers were originally developed by the same now-defunct third party, and have been included in Windows for decades. These driver removals will pass unnoticed for most people, but you might find active modems still in a few contexts, including some industrial control systems.”

According to Barnett, two questions remain: How many more legacy modem drivers are still present on a fully-patched Windows asset; and how many more elevation-to-SYSTEM vulnerabilities will emerge from them before Microsoft cuts off attackers who have been enjoying “living off the land[line] by exploiting an entire class of dusty old device drivers?”

“Although Microsoft doesn’t claim evidence of exploitation for CVE-2023-31096, the relevant 2023 write-up and the 2025 removal of the other Agere modem driver have provided two strong signals for anyone looking for Windows exploits in the meantime,” Barnett said. “In case you were wondering, there is no need to have a modem connected; the mere presence of the driver is enough to render an asset vulnerable.”

Immersive, Ivanti and Rapid7 all called attention to CVE-2026-21265, which is a critical Security Feature Bypass vulnerability affecting Windows Secure Boot. This security feature is designed to protect against threats like rootkits and bootkits, and it relies on a set of certificates that are set to expire in June 2026 and October 2026. Once these 2011 certificates expire, Windows devices that do not have the new 2023 certificates can no longer receive Secure Boot security fixes.

Barnett cautioned that when updating the bootloader and BIOS, it is essential to prepare fully ahead of time for the specific OS and BIOS combination you’re working with, since incorrect remediation steps can lead to an unbootable system.

“Fifteen years is a very long time indeed in information security, but the clock is running out on the Microsoft root certificates which have been signing essentially everything in the Secure Boot ecosystem since the days of Stuxnet,” Barnett said. “Microsoft issued replacement certificates back in 2023, alongside CVE-2023-24932 which covered relevant Windows patches as well as subsequent steps to remediate the Secure Boot bypass exploited by the BlackLotus bootkit.”

Goettl noted that Mozilla has released updates for Firefox and Firefox ESR resolving a total of 34 vulnerabilities, two of which are suspected to be exploited (CVE-2026-0891 and CVE-2026-0892). Both are resolved in Firefox 147 (MFSA2026-01) and CVE-2026-0891 is resolved in Firefox ESR 140.7 (MFSA2026-03).

“Expect Google Chrome and Microsoft Edge updates this week in addition to a high severity vulnerability in Chrome WebView that was resolved in the January 6 Chrome update (CVE-2026-0628),” Goettl said.

As ever, the SANS Internet Storm Center has a per-patch breakdown by severity and urgency. Windows admins should keep an eye on askwoody.com for any news about patches that don’t quite play nice with everything. If you experience any issues related installing January’s patches, please drop a line in the comments below.

Our first story of 2026 revealed how a destructive new botnet called Kimwolf has infected more than two million devices by mass-compromising a vast number of unofficial Android TV streaming boxes. Today, we’ll dig through digital clues left behind by the hackers, network operators and services that appear to have benefitted from Kimwolf’s spread.

On Dec. 17, 2025, the Chinese security firm XLab published a deep dive on Kimwolf, which forces infected devices to participate in distributed denial-of-service (DDoS) attacks and to relay abusive and malicious Internet traffic for so-called “residential proxy” services.

The software that turns one’s device into a residential proxy is often quietly bundled with mobile apps and games. Kimwolf specifically targeted residential proxy software that is factory installed on more than a thousand different models of unsanctioned Android TV streaming devices. Very quickly, the residential proxy’s Internet address starts funneling traffic that is linked to ad fraud, account takeover attempts and mass content scraping.

The XLab report explained its researchers found “definitive evidence” that the same cybercriminal actors and infrastructure were used to deploy both Kimwolf and the Aisuru botnet — an earlier version of Kimwolf that also enslaved devices for use in DDoS attacks and proxy services.

XLab said it suspected since October that Kimwolf and Aisuru had the same author(s) and operators, based in part on shared code changes over time. But it said those suspicions were confirmed on December 8 when it witnessed both botnet strains being distributed by the same Internet address at 93.95.112[.]59.

RESI RACK

Public records show the Internet address range flagged by XLab is assigned to Lehi, Utah-based Resi Rack LLC. Resi Rack’s website bills the company as a “Premium Game Server Hosting Provider.” Meanwhile, Resi Rack’s ads on the Internet moneymaking forum BlackHatWorld refer to it as a “Premium Residential Proxy Hosting and Proxy Software Solutions Company.”

Resi Rack co-founder Cassidy Hales told KrebsOnSecurity his company received a notification on December 10 about Kimwolf using their network “that detailed what was being done by one of our customers leasing our servers.”

“When we received this email we took care of this issue immediately,” Hales wrote in response to an email requesting comment. “This is something we are very disappointed is now associated with our name and this was not the intention of our company whatsoever.”

The Resi Rack Internet address cited by XLab on December 8 came onto KrebsOnSecurity’s radar more than two weeks before that. Benjamin Brundage is founder of Synthient, a startup that tracks proxy services. In late October 2025, Brundage shared that the people selling various proxy services which benefitted from the Aisuru and Kimwolf botnets were doing so at a new Discord server called resi[.]to.

When KrebsOnSecurity joined the resi[.]to Discord channel in late October as a silent lurker, the server had fewer than 150 members, including “Shox” — the nickname used by Resi Rack’s co-founder Mr. Hales — and his business partner “Linus,” who did not respond to requests for comment.

Other members of the resi[.]to Discord channel would periodically post new IP addresses that were responsible for proxying traffic over the Kimwolf botnet. As the screenshot from resi[.]to above shows, that Resi Rack Internet address flagged by XLab was used by Kimwolf to direct proxy traffic as far back as November 24, if not earlier. All told, Synthient said it tracked at least seven static Resi Rack IP addresses connected to Kimwolf proxy infrastructure between October and December 2025.

Neither of Resi Rack’s co-owners responded to follow-up questions. Both have been active in selling proxy services via Discord for nearly two years. According to a review of Discord messages indexed by the cyber intelligence firm Flashpoint, Shox and Linus spent much of 2024 selling static “ISP proxies” by routing various Internet address blocks at major U.S. Internet service providers.

In February 2025, AT&T announced that effective July 31, 2025, it would no longer originate routes for network blocks that are not owned and managed by AT&T (other major ISPs have since made similar moves). Less than a month later, Shox and Linus told customers they would soon cease offering static ISP proxies as a result of these policy changes.

DORT & SNOW

The stated owner of the resi[.]to Discord server went by the abbreviated username “D.” That initial appears to be short for the hacker handle “Dort,” a name that was invoked frequently throughout these Discord chats.

This “Dort” nickname came up in KrebsOnSecurity’s recent conversations with “Forky,” a Brazilian man who acknowledged being involved in the marketing of the Aisuru botnet at its inception in late 2024. But Forky vehemently denied having anything to do with a series of massive and record-smashing DDoS attacks in the latter half of 2025 that were blamed on Aisuru, saying the botnet by that point had been taken over by rivals.

Forky asserts that Dort is a resident of Canada and one of at least two individuals currently in control of the Aisuru/Kimwolf botnet. The other individual Forky named as an Aisuru/Kimwolf botmaster goes by the nickname “Snow.”

On January 2 — just hours after our story on Kimwolf was published — the historical chat records on resi[.]to were erased without warning and replaced by a profanity-laced message for Synthient’s founder. Minutes after that, the entire server disappeared.

Later that same day, several of the more active members of the now-defunct resi[.]to Discord server moved to a Telegram channel where they posted Brundage’s personal information, and generally complained about being unable to find reliable “bulletproof” hosting for their botnet.

Hilariously, a user by the name “Richard Remington” briefly appeared in the group’s Telegram server to post a crude “Happy New Year” sketch that claims Dort and Snow are now in control of 3.5 million devices infected by Aisuru and/or Kimwolf. Richard Remington’s Telegram account has since been deleted, but it previously stated its owner operates a website that caters to DDoS-for-hire or “stresser” services seeking to test their firepower.

BYTECONNECT, PLAINPROXIES, AND 3XK TECH

Reports from both Synthient and XLab found that Kimwolf was used to deploy programs that turned infected systems into Internet traffic relays for multiple residential proxy services. Among those was a component that installed a software development kit (SDK) called ByteConnect, which is distributed by a provider known as Plainproxies.

ByteConnect says it specializes in “monetizing apps ethically and free,” while Plainproxies advertises the ability to provide content scraping companies with “unlimited” proxy pools. However, Synthient said that upon connecting to ByteConnect’s SDK they instead observed a mass influx of credential-stuffing attacks targeting email servers and popular online websites.

A search on LinkedIn finds the CEO of Plainproxies is Friedrich Kraft, whose resume says he is co-founder of ByteConnect Ltd. Public Internet routing records show Mr. Kraft also operates a hosting firm in Germany called 3XK Tech GmbH. Mr. Kraft did not respond to repeated requests for an interview.

In July 2025, Cloudflare reported that 3XK Tech (a.k.a. Drei-K-Tech) had become the Internet’s largest source of application-layer DDoS attacks. In November 2025, the security firm GreyNoise Intelligence found that Internet addresses on 3XK Tech were responsible for roughly three-quarters of the Internet scanning being done at the time for a newly discovered and critical vulnerability in security products made by Palo Alto Networks.

LinkedIn has a profile for another Plainproxies employee, Julia Levi, who is listed as co-founder of ByteConnect. Ms. Levi did not respond to requests for comment. Her resume says she previously worked for two major proxy providers: Netnut Proxy Network, and Bright Data.

Synthient likewise said Plainproxies ignored their outreach, noting that the Byteconnect SDK continues to remain active on devices compromised by Kimwolf.

MASKIFY

Synthient’s January 2 report said another proxy provider heavily involved in the sale of Kimwolf proxies was Maskify, which currently advertises on multiple cybercrime forums that it has more than six million residential Internet addresses for rent.

Maskify prices its service at a rate of 30 cents per gigabyte of data relayed through their proxies. According to Synthient, that price range is insanely low and is far cheaper than any other proxy provider in business today.

“Synthient’s Research Team received screenshots from other proxy providers showing key Kimwolf actors attempting to offload proxy bandwidth in exchange for upfront cash,” the Synthient report noted. “This approach likely helped fuel early development, with associated members spending earnings on infrastructure and outsourced development tasks. Please note that resellers know precisely what they are selling; proxies at these prices are not ethically sourced.”

Maskify did not respond to requests for comment.

BOTMASTERS LASH OUT

Hours after our first Kimwolf story was published last week, the resi[.]to Discord server vanished, Synthient’s website was hit with a DDoS attack, and the Kimwolf botmasters took to doxing Brundage via their botnet.

The harassing messages appeared as text records uploaded to the Ethereum Name Service (ENS), a distributed system for supporting smart contracts deployed on the Ethereum blockchain. As documented by XLab, in mid-December the Kimwolf operators upgraded their infrastructure and began using ENS to better withstand the near-constant takedown efforts targeting the botnet’s control servers.

By telling infected systems to seek out the Kimwolf control servers via ENS, even if the servers that the botmasters use to control the botnet are taken down the attacker only needs to update the ENS text record to reflect the new Internet address of the control server, and the infected devices will immediately know where to look for further instructions.

“This channel itself relies on the decentralized nature of blockchain, unregulated by Ethereum or other blockchain operators, and cannot be blocked,” XLab wrote.

The text records included in Kimwolf’s ENS instructions can also feature short messages, such as those that carried Brundage’s personal information. Other ENS text records associated with Kimwolf offered some sage advice: “If flagged, we encourage the TV box to be destroyed.”

Both Synthient and XLabs say Kimwolf targets a vast number of Android TV streaming box models, all of which have zero security protections, and many of which ship with proxy malware built in. Generally speaking, if you can send a data packet to one of these devices you can also seize administrative control over it.

If you own a TV box that matches one of these model names and/or numbers, please just rip it out of your network. If you encounter one of these devices on the network of a family member or friend, send them a link to this story (or to our January 2 story on Kimwolf) and explain that it’s not worth the potential hassle and harm created by keeping them plugged in.