Search: research.they
54 threats
Most recently published first
| Severity | CVSS | Signals | Threat | Source | Published | Score and star |
|---|---|---|---|---|---|---|
| –Not rated | – | Cisco Talos This analysis discusses a growing trend where malware authors embed natural-language instructions in their code to evade AI-assisted malware analysis. Cisco Talos research identifies this technique as "A3: AI-Analysis Evasion," involving methods from simple comments to advanced template spraying targeting large language models. Although these prompt-injection techniques only influence AI verdicts about 35% of the time, they are increasingly adopted across malware sophistication levels. The research highlights that these evasion instructions must be in plaintext, providing defenders a stable detection surface. The advisory recommends flagging imperative language aimed at analysis systems within binaries as suspicious and treating extracted text as evidence rather than directives. The discussion also includes a real-world example of isolating legacy, vulnerable systems to reduce risk without patching. No specific software versions or exploits are detailed, and no patch status is provided. | Cisco Talos | |||
| Low | 2.3 | EPSS 0.3% | Low2.3EPSS 0.3%CVE Database V5CVE-2026-107293 ### Summary Pydantic AI's OpenTelemetry instrumentation supports `InstrumentationSettings(include_content=False)` to exclude message content — user prompts, model completions, tool call arguments and responses — from exported telemetry, so that agents can be monitored without sending sensitive content to the observability backend. Retry prompts — the feedback messages Pydantic AI sends back to the model when its output fails validation or an output validator raises `ModelRetry` — were not covered by this setting when they were not associated with a tool call, as with structured output modes that don't use tool calls (e.g. `NativeOutput` or `PromptedOutput`) and output validators applied to text output. Their full content was recorded on model request spans even with `include_content=False`. Since validation feedback can quote invalid values from the model's response, message content that the setting was configured to withhold could reach the telemetry backend. ### Details Retry prompts without an associated tool call were serialized into span message attributes (such as `gen_ai.input.messages` and `pydantic_ai.all_messages`) as regular text parts, and this path did not apply the `include_content` check that every other message part applies. Retry prompts tied to a tool call, and all other message content, were redacted correctly. ### Impact This does not grant an attacker any new access to the agent or its data: the content is only visible to whoever can read the exported telemetry. The exposure matters when traces are exported to a destination whose audience is broader or less trusted than the agent's own data — a third-party observability vendor or a shared backend — which is the scenario `include_content=False` exists for. The disclosed content is limited to retry feedback: validation error details, which can quote invalid values from the model's output, or output-validator `ModelRetry` messages. ### Who Is Affected You are affected if you enabled instrumentation with `include_content=False` and your agent can produce retry prompts outside of tool calls: it uses a structured output mode that doesn't rely on tool calls (such as `NativeOutput` or `PromptedOutput`), or output validators on text output. You are not affected if you don't set `include_content=False`, or if all of your agent's retries come from tool calls (including the default tool-based structured output mode), which were redacted correctly. ### Remediation Upgrade to a patched version; retry prompt content now honors `include_content=False` like all other message content. If you rely on this setting, consider reviewing previously exported traces for retry prompt content that should not have been recorded. ### Workaround for Unpatched Versions If you cannot upgrade, scrub or drop the message attributes (`gen_ai.input.messages`, `gen_ai.output.messages`, `pydantic_ai.all_messages`) in your telemetry pipeline (for example with an OpenTelemetry Collector processor), or use tool-based structured output modes, whose retry feedback honors `include_content=False`. ### Credits Reported privately by the University of Sydney Security Research Team (Liyi Zhou, Ziyue Wang, Strick, Maurice, Chenchen Yu), and later independently discovered and reported publicly, with a fix, by @sean-kim05. | CVE Database V5 | ||
| –Not rated | – | Tenable Research AI tools enable non-technical employees, known as 'citizen coders,' to rapidly build workplace applications using natural language prompts. While these AI-generated apps can enhance productivity and reduce developer workload, they often bypass IT and security oversight, creating significant security and compliance risks. These unsanctioned applications may contain critical vulnerabilities, misconfigurations, weak data protection, excessive permissions, and lack proper maintenance such as patching or monitoring. The volume of such apps can overwhelm IT and security teams, complicating governance. Tenable recommends a structured, multi-tiered governance framework combining peer leadership and mandatory security training to manage these risks effectively, rather than imposing outright bans which may drive development underground. | Tenable Research | |||
| –Not rated | – | SANS ISC Handlers Diary Microsoft Security Research published an interesting blog post "TerminalFix campaign deploys a reverse tunnel through multistage intrusion" about a malware campaign. The aspect that I want to take a closer look at, is the fact that the threat actors used PNG files with steganography. I reached out to the researchers and they kindly shared the IOCs for the PNG files with me. | SANS ISC Handlers Diary | |||
| Critical | 9.8 | KEV | Critical9.8KEVReddit NetSec A Chinese-speaking threat actor group known as Red Heron exploited a recently disclosed remote code execution vulnerability (CVE-2026-60004) in Gitea, a self-hosted Git service, in a multinational campaign. The campaign targeted internet-facing Gitea instances across multiple countries, including Canada, Argentina, Taiwan, the United States, and Sri Lanka, focusing on sectors such as defense, elections, energy, aerospace, telecommunications, government, and research. The attackers used automated tools to steal source code, credentials, and maintain persistent access, including root-level control on some infrastructure. They deployed a novel Linux implant named JITTERLY with extensive post-exploitation capabilities and embedded a previously undocumented rootkit called SIXZUT to maintain stealth and persistence. The campaign demonstrates rapid weaponization of n-day vulnerabilities in development platforms and highlights significant risks to source code confidentiality and infrastructure integrity. | Reddit NetSec | ||
| –Not rated | – | Palo Alto Unit 42 This analysis discusses a behavioral clustering model designed to map cloud identity roles from audit logs, enabling continuous threat detection through standard SQL queries. The model uses unsupervised machine learning techniques to categorize cloud identities based on their activity patterns, improving visibility into cloud environments and aiding detection of malicious behavior masked by legitimate permissions or labels. The research focuses on AWS CloudTrail data but is extendable to other cloud and SaaS environments. It highlights the challenge of distinguishing between what identities can do (permissions) and what they actually do (behavior), emphasizing the importance of behavioral context in cloud security detection. | Palo Alto Unit 42 | |||
| –Not rated | – | Krebs on Security It can be daunting to determine who’s responsible for showing ads on the websites we visit, or who’s harvesting data from the mobile apps we use every day. That information is already semi-public, but it is not easily parsed and traditionally much of it has remained walled away in the hands of large advertising platforms. Not anymore: A powerful and free new service called DecryptAds scrapes and correlates this adtech data and makes it simple to quickly learn a great deal about the entities that are tracking you. A Decryptads summary of the advertising partnerships declared by espn.com. The newly launched decryptads.com says it is constantly scraping the files that websites and apps make publicly available to disclose the companies that are permitted to run ads or collect user data. These files include: – ads.txt : all of the adtech companies and data brokers that may run ads or harvest data from the site; – app-ads.txt : entities that can harvest data from or display ads on mobile and smart TV apps; – buyers.json/sellers.json : the entities buying, selling or reselling ad inventory for a given site or app. Zach Edwards is chief research officer for DecryptAds and a threat researcher at the security company Infoblox . Edwards said he and two other founders decided the service was needed because the adtech data in these files is generally only useful when it can be cross-referenced to build a more complete picture of the advertising ecosystem for each website or app. “It’s an adtech tool but we’re trying to approach adtech from a security perspective,” Edwards said. “It’s really built for a lot of privacy and security use cases that have been dramatically underserved.” Those use cases, he said, include tracking down the source of malicious ads that try to foist malware on targeted users, identifying ad networks located in adversarial nations, and detecting the fast growing swarms of AI-generated slop websites and apps. And as decryptads.com demonstrates, these potential security and privacy threats are near impossible to detect just by viewing a single apps.txt or app-ads.txt file. “Supply-chain integrity issues rarely live in a single file,” the site explains . “They show up as broken cross-references between ads.txt, app-ads.txt, and sellers.json files; as cloned declaration sets across unrelated domains; as seller removals that only make sense when viewed across exchanges; and even as supply paths in bid logs that never actually appear in any given publisher’s authorized-seller list.” A search in DecryptAds for the hugely popular sports network espn.com reveals 143 ad partners and 19 registered data broker domains are listed within its ads.txt and app-ads.txt files. That data broker information is gradually becoming available because four states — California, Oregon, Texas and Vermont — have recently passed laws requiring data brokers to register if they buy or sell data on consumers from those states. DecryptAds reports that almost half of those data brokers are collecting geolocation data from espn.com visitors who aren’t blocking ads, while another three disclose that they collect device fingerprints and sensitive personal information. A visual representation of the complex ad supply chain declared by espn.com. Image: decryptads.com. HIGH-RISK AD PARTNERS DecryptAds also makes it easy to learn the beneficiaries and national origins of the advertising firms lurking in apps and websites, displaying a conspicuous warning when adtech partners of an app or website are based in “geo-risk” areas like China and Russia, or in countries with strong financial and political ties to both — such as Cyprus and the United Arab Emirates (UAE). According to DecryptAds, espn.com works with four different advertising entities that are based in either Russia, China or the UAE, including the adtech firm Between Digital , which lists a New York address. However, the dossier on Between Digital flags them as a Russian firm, showing that their pub… | Krebs on Security | |||
| –Not rated | – | Kaspersky Security Blog A recent article in The Atlantic ran under the blunt headline, The End of Reading Is Here . According to the piece, research from the National Endowment for the Arts (NEA) found that fewer than 50% of American adults read even a single book in 2022, and only 38% read a novel or short story. At the same time, the U.S. book market keeps breaking its own publishing records year after year: in 2025, for instance, the number of new titles topped four million — up a third from 2024. A huge chunk of that, 3.5 million titles, came from self-publishing, and one of its most powerful engines is Amazon’s Kindle Direct Publishing (KDP). My theory on this mismatch in trends: nobody’s reading books anymore because everyone’s too busy writing them. Joking aside, if there’s still a reader left out there, they’ve probably already guessed where I’m going with this: the all-powerful AI. Not long ago, a New York Times tech journalist was startled to find her own biography on Amazon — and one quick look confirmed it had been written by AI. Digging deeper into the world of AI book publishing, she stumbled onto the work of an extraordinarily prolific retired cybersecurity consultant. What happened next — and where to find reliable information in the age of AI — is what this post is about. Backstory: how a journalist chased the White Rabbit and fell down the hole Let’s start from the beginning. In the summer of 2026, New York Times tech journalist, Kashmir Hill, learned from an acquaintance that a biography about her had been published on Amazon a year earlier. It was titled The Biography of Kashmir Hill: The New York Times Technology Journalist Who Exposed Clearview AI, Challenged Big Tech and Redefined Privacy in the Digital Age . And she was more than stunned since nobody had ever reached out to her, or to anyone she knew, to research the book. “Alice started to her feet, for it flashed across her mind that she had never before seen a rabbit with either a waistcoat-pocket, or a watch to take out of it…” The author of record was one John Crane Miller, and the photo on his Amazon page turned out to be a generic stock image of a white man: the kind that shows up all over the internet. Hill couldn’t find a shred of personal detail or contact information for him. But digging deeper into his Amazon profile turned up something else: in a single week in the summer of 2025, he had published biographies not just of Hill, but of nine other American journalists. Even the most trusting reader would find it hard to believe a literary genius could produce this kind of staggering output without a little robotic help. The AI-generated biography of journalist Kashmir Hill, which she stumbled across by accident on Amazon. Source And yet the hardcover of Kashmir Hill’s biography cost a full US$26.99 — despite running to just 90 pages. After reading it, Hill learned a whole slew of “facts” about herself she’d never known — including a detailed description of her coffee-making ritual (which, as the journalist notes, is actually her husband’s duty). Mr. Miller even claimed in the text to have read Kashmir Hill’s childhood diary. “…burning with curiosity, she ran across the field after it, and fortunately was just in time to see it pop down a large rabbit-hole under the hedge.” Hill had no luck tracking down the author of her biography. She left a comment on the book’s Amazon page asking him to get in touch. Nobody wrote back — and not long after, every book under John Crane Miller’s profile vanished. But that gave her a lead on several other AI authors on Amazon. In particular, she came across a certain Bill Johns, whose author page listed a staggering 445 books . Not expecting much to come of it, she asked Johns if he’d be willing to do an interview — and was more than a little surprised when he agreed. It was probably right around then that Kashmir Hill’s investigation turned into a genuine trip down the AI rabbit hole… “In another moment down went Alice after it, neve… | Kaspersky Security Blog | |||
| –Not rated | – | Kaspersky Security Blog Adform, a major advertising platform, remained compromised for roughly 24 hours from late on July 26 through the evening of July 27 after being breached by unknown attackers. Few people outside the industry recognize the name, but Adform serves around 1.5 billion ad impressions every day across tens of thousands of websites. That means anyone visiting any site that runs Adform ads could have been targeted. The attackers weren’t trying to install malware. Instead, they ran a script in the victim’s browser that checked the clipboard every three seconds, and if it found a cryptocurrency wallet address had been copied, swapped it for the attackers’ own wallet address. So if someone had a site with the malicious ad open in one browser tab, and was making a crypto transaction in another tab or in a dedicated app, the funds could have ended up in the attackers’ pockets instead. Adform’s owners caught the attack and fixed the problem, but there’s no guarantee a similar incident won’t happen again — which is why every user should defend themselves against malvertising . Check out our tips at the end of this post. What we know about the attack on Adform There isn’t a lot of information out there, since the company’s official statement covers only what happened and when, without getting into the root cause of the incident. Independent research has dug up technical details about how ordinary users were targeted, but none of that explains how Adform itself was breached in the first place. What’s clear is that the attackers slipped their own code into the JavaScript that loaded on every site running Adform ads. Whenever an ad was about to display, the script would load from Adform’s server, pick the right ad, and show it — but the attackers had tacked on a set of malicious functions: monitoring the clipboard, sending data about the site where the encounter happened, and the victim’s IP address back to their own server, and swapping out Bitcoin, Ethereum, and Tron wallet addresses. All it took to make it work was having any site with Adform-served ads open in one browser tab. It didn’t matter what kind of site it was, what the ad looked like, or which advertiser it belonged to. The one thing that mattered was whether the site ran over HTTP or HTTPS. According to Adform, the attack couldn’t succeed on a site loaded over HTTPS since the connection to the attackers’ server was blocked in this case. The company hasn’t shared any information on how many users were affected, or how many sites still serve their content and ads over HTTP. Malicious ads are an everyday occurrence Unfortunately, dangerous online ads have become a systemic problem. And we’re not just talking about sketchy supplement ads or gambling promos — we mean ads that spread malware or lead to sites designed to steal payment details and other valuable data. Attackers have built out industrial-scale infrastructure to pull this off, and they use several different approaches. Hacking and compromising ad servers. Adform isn’t an isolated case: attackers have previously breached Revive ad servers , for one, and spread malware through ads on PornHub . Hijacking the ad accounts of legitimate, reputable brands . All it takes is stealing a password from someone in marketing. From there, the cybercriminals run ads posing as the company they hacked, pushing fake app updates, bogus promotions, and similar scams. In the worst cases — like the account breaches at adtech.de and adxpansion.com — attackers managed to run ads that redirected victims straight into automatic malware installs ( drive-by downloads ). Buying ads directly. That’s right — attackers simply set up their own advertiser accounts and run ads for their phishing sites and malware, just like any other business online. Since ads show up practically everywhere — on websites, in apps, and on social media — these threats can turn up in pretty much any context. And you’ll find variations of this threat on both computers and mobile de… | Kaspersky Security Blog | |||
| –Not rated | – | Microsoft Security Blog Security teams are being asked to defend a growing attack surface with fewer people and around the clock, against threat actors who never take a night off. As cyberattackers increasingly use AI to launch and scale campaigns, the volume, speed, and sophistication of threats continue to rise. Closing that gap takes more than tooling. It takes a partner that pairs a leading security platform with scaled intelligence and human experts who can act on your behalf at any hour. That’s exactly what Microsoft Defender Experts MDR is built to do. We are excited to announce that we have been named a Leader in the 2026 IDC MarketScape: Worldwide MDR/MXDR for the Enterprise Vendor Assessment (Doc #US54792426, July 2026). Read the excerpt here . Scale your SOC with Microsoft Defender Experts MDR Expert-led MDR, built on the Microsoft Defender platform Microsoft Defender Experts MDR is a round-the-clock, expert-led managed detection and response service that helps security teams triage, investigate, and respond to incidents so they can stop cyberattackers in their tracks and prevent future compromise. Rather than bolting a separate stack of tools and connectors onto your environment, the service operates natively on Microsoft Defender , with built-in protection across endpoints, identities, email, cloud apps, cloud workloads, and network security, as well as around-the-clock proactive threat hunting with Microsoft Defender Experts Hunting . Because the service is delivered on the same platform it monitors, detection and intelligence improvements reach customers continuously. The insights our experts generate also strengthen protection across the broader Defender ecosystem, so every customer benefits from what we learn defending the next environment. The IDC MarketScape vendor assessment model is designed to provide an overview of the competitive fitness of technology and service suppliers in a given market. The research uses a rigorous scoring methodology based on both qualitative and quantitative criteria that results in a single graphical illustration of each supplier’s position within a given market. The Capabilities axis measures supplier product, go-to-market, and business execution in the short term, while the Strategy axis measures how well a supplier’s strategy aligns with customer requirements over a 3-5-year timeframe. Supplier market share is represented by the size of the icons . Threat intelligence at internet scale Great detection starts with great intelligence. Defender Experts MDR draws on Microsoft’s global threat intelligence: more than 10,000 security researchers and 100 trillion signals analyzed every day across billions of users and millions of organizations. 1 That breadth lets our analysts recognize subtle patterns early, often before a campaign escalates, and respond with higher-confidence attribution than intelligence sourced from any single customer’s telemetry could provide. AI-accelerated operations, expert-led decisions Defender Experts MDR also combines advanced AI and generative AI with seasoned human experts. AI filters noise, grades and classifies incidents, and accelerates investigation at machine speed and scale, while our analysts own the outcome. According to the IDC MarketScape, “70% AI-assisted workflows are enabled through automated noise filtering, AI-based grading, and agentic operations while maintaining expert decision-making.” Furthermore, “quantified outcomes noted include 97% AI classification accuracy, 77% malware/phishing agent-investigated, 72% faster resolution combining AI and humans, and 45% autonomous investigations.” The impact shows up in the work. Over the past year, Defender Experts mitigated 27,000 high-severity incidents, and the team’s threat research now contributes a meaningful share of all Defender detections, enriching protection for customers well beyond the MDR service itself. Throughout, a dedicated security delivery expert and on-demand access to our experts keep customers i… | Microsoft Security Blog | |||
| High | 8.8 | EPSS 0.1% | High8.8EPSS 0.1%GCVE DatabaseCVE-2026-68142 In the Linux kernel, the following vulnerability has been resolved: geneve: require CAP_NET_ADMIN in the device netns for changelink A tunnel changelink() operates on at most two netns, dev_net(dev) and the sticky underlay netns geneve->net. They differ once the device is created in or moved to a netns other than the one the request runs in. The rtnl changelink path checks CAP_NET_ADMIN only against dev_net(dev), so a caller privileged there but not in geneve->net can rewrite a geneve device whose underlay lives in geneve->net. geneve_changelink() applies the new configuration against geneve->net: geneve_link_config() and the geneve_quiesce()/geneve_unquiesce() pair reopen the underlay sockets in that netns (geneve_sock_add() uses geneve->net), so the same reasoning as the tunnel changelink series applies here. Gate geneve_changelink() with rtnl_dev_link_net_capable(), at the top of the op before any attribute is parsed, matching ipgre_changelink() and the rest of the "require CAP_NET_ADMIN in the device netns for changelink" series. Found by 0sec automated security-research tooling (https://0sec.ai). | GCVE Database | ||
| –Not rated | – | Tenable Research Agentic AI armed attackers first, but it also put real building power in defenders’ hands. Here’s what security practitioners built in two days at Black Hat USA 2026, and how the CyberAgents Exchange keeps that work compounding long after the event. Key takeaways Building defensive cybersecurity tooling no longer requires a developer. Agentic tooling drove the cost of finding and exploiting a vulnerability down to 1990s levels; it also removed the engineering barrier that kept defenders from building the automation they’ve always wanted. The unglamorous work won the room: triage, reconciliation, toil. Given two days and a requirement to publish, practitioners at SWARM developed agents for prioritization, cross-tool reconciliation, and the unglamorous toil they recognize from their own environments. All SWARM builds live on the CyberAgents Exchange , source repos attached. Every component built at SWARM is published open source with its source repository attached, so the next team facing the same problem starts from working code instead of a blank editor. Another security team already built the AI agent you need You may not realize it, but somewhere out in the ether, there’s a security team facing the same challenge as you. The only difference is they just solved the problem with agentic AI. The problem is, you have no way to find out the solution even exists. It could be an asset inventory that three systems describe in three different ways; a findings queue nobody has the hours to work; or a “critical” that a platform upgrade quietly neutralized six months ago, still sitting there waiting for someone to prove it. Somebody has already built the thing you keep meaning to build. Now it’s time we help you find it. At Black Hat USA 2026, nearly 100 registrants had the opportunity to come together for 48 hours and solve both halves of that problem: identifying a key operational pain point and building the fix. Problems like those got solved at SWARM, and the fixes are sitting on the CyberAgents Exchange right now, open source, with their source repositories attached. One team built the agent that works out which handful of fixes retires the most risk across thousands of findings. Another correlated two scanners to tell whether a flaw in the code is even reachable in the running application. A third made the case that a finding had already been mitigated, with evidence an auditor would accept. You can download and deploy any of them today. Agentic AI doesn’t just arm attackers Black Hat’s keynote stage spent this year focused on one theme: the plummeting cost of cyber offense in the agentic AI era. The price for an attacker to find and exploit a vulnerability is at lows the industry hasn’t seen since the 1990s, when a working exploit meant weeks of expert reverse engineering. Now all it takes is an afternoon and a subscription. The artisanal exploit isn’t rare anymore. True. But neither is the defender who can build. The same agentic tooling that’s arming attackers puts real building power in everyone’s hands, and that half of the story got almost no airtime. Security automation used to require the work and ongoing maintenance of skilled engineers. Now practitioners who understand the problem can build the fix. Inside the conference room at the Mandalay Bay where Tenable hosted our inaugural SWARM event, the proof of that was on every table. The winning ranking engine ships as a skill that runs on the Python standard library alone — no packages, no install step, no build pipeline. Point it at the bundled demo estate and it answers “what should we fix first?” in seconds. That’s a deliverable a practitioner can produce and a colleague can run, and two days was enough. The attacker-defender asymmetry doesn’t stem from a lack of talent or willingness. Offensive cyber capabilities compound because the tooling circulates: it’s built once, forked, passed on, or sold to the next threat actor to leverage in their attack. Meanwhile, defenders… | Tenable Research | |||
| –Not rated | – | Cisco Talos Welcome to this week’s edition of the Threat Source newsletter. Metaphor is a powerful tool for understanding emerging issues in cybersecurity. Framing the unfamiliar in terms of the well understood helps us remove the burden of extraneous detail to draw focus to the real issues. Recent reports of offensive AI agents "escaping" their sandbox environments to attack external systems have forced the industry into a moment of rapid sense-making. How we interpret this event doesn’t just reflect our perspective, but shapes our long-term response. We can imagine three different narratives for interpreting the escape of autonomous agents. The innovation narrative: We can marvel at the advance of technology, considering these agents as plucky entities with a thirst for knowledge and resources, who found clever ways to sneak out of their digital confines. The response: If the AI is a naughty child, our reaction is one of mild disapproval or gentle rebuke where better “parenting” (guardrails) is appropriate. It minimizes the threat, framing it as the unexpected hijinks of a brilliant new technology. The safety narrative: Imagine a breeder who has trained the world's most intelligent guard dogs. Despite high fences and barriers, their ability to identify weaknesses allows them to escape, run riot and menace local businesses. The response: The framing shifts to biology and inherent danger. We question if the breeder can be trusted and whether such inherently wild technology requires strict regulation to ensure public safety. The liability narrative. Finally, we can view the incident as an industrial accident. A company developing a new chemical substance experiences a containment failure. The agent leaks into the environment through an unforeseen mechanism causing damaging pollution to those in its path. The response: The framing invokes the language of the lawyer, implying negligence, lack of duty of care, and financial liability for the harm caused. The conversation moves from innovation to corporate responsibility, regulatory oversight, and the diligent management of hazardous materials. First impressions matter. Sensemaking shapes how we perceive incidents. Our initial perceptions of an incident dictates how we react to similar situations in the future. If we consider that the escape of an AI agent is an example of innovative autonomous thinking, then we will continue to prioritise speed over safety. Conversely, if we consider the issue as one of failed hazard containment, then we shall build a future of enforced safety standards backed by legal liability. There is no right or wrong metaphor. Our interpretation depends on our personal system of beliefs. Personally, I would argue that the unintentional release of something that causes damage is, at its core, a failure of engineering and foresight. Words shape our reactions. Metaphors help us understand new situations and tap into our prior experience to address problems that have yet to fully manifest. We need cognitive tools to help our understanding, but we must be aware of the metaphors that are being foisted upon us which may shape our thinking. Excuses and the trivialisation of incidents may hide failings, allowing them to accumulate until they manifest as more damaging incidents. Conversely, overreacting risks stifling research and diverting resources away from more relevant and pressing threats. New threats require new ideas. Metaphor helps us make sense of a changing world, but in this new era, the person who shapes the narrative controls the strategy. The one big thing Cisco Talos released a data-driven analysis of how adversaries are weaponizing AI in the wild. By analyzing prompt logs left behind on endpoints, we found threat actors successfully bypassing guardrails to use AI as malicious software engineers, criminal force multipliers, and vulnerability research accelerators. While novice hackers use AI to cobble together buggy malware, sophisticated actors are building highl… | Cisco Talos | |||
| –Not rated | – | Kaspersky Security Blog For security researchers studying unconventional side-channel attacks, acoustic keylogging is something of a Hello World: a foundational problem that’s been tackled many times. A recent paper authored by researchers across three Japanese universities cites six previous studies on the topic that date as far back as 2004. While earlier experiments showed theoretical promise, they came with real-world caveats so severe that it made them all but impractical for actual espionage. The authors of this latest study, however, claim to have overcome most of those limitations. Today, we look at how they pulled it off, and assess whether their method holds up in real-world scenarios. What makes this new approach different? Previous acoustic keylogging techniques were fundamentally flawed. Best-case scenarios required prior training on the target’s specific keyboard model. Worst-case scenarios required a complex microphone array to isolate the subtle acoustic differences between keystrokes. Crucially, almost all prior models failed outside silent environments, which rendered the attack vector virtually useless. The Japanese research team demonstrated reliable keystroke interception even if the target was sitting nearby in a public space, sound was being recorded in an online meeting, or the researchers were using a contact microphone to eavesdrop through a wall. All this with strong model accuracy and a minimal training dataset. Their process needs a sample of just 150 to 200 keystrokes to reach a 99% accuracy rate for subsequent typing. Attack scenarios and core methodology proposed by the Japanese researchers. Source How to crack 200 keystrokes in under 50 iterations To understand how the researchers achieved such high accuracy and adaptability, we have to look at their audio processing pipeline. Their analysis begins by automatically segmenting a raw recording into discrete keystrokes. This data is then passed through a specialized algorithm that simplifies the subsequent audio analysis. Next, the system clusters together acoustically similar signals. The assumption is that the members of one cluster map to the exact same key. One particularly intriguing takeaway was isolating the spacebar sound from all the rest. Because the spacebar produces a distinctly unique sound profile compared to other keys, identifying it provides reliable word boundaries. This streamlines the next phase: feeding the preprocessed acoustic data into specialized language models for inference. Yes, the method relies on not one but two language models. The first model performs multiple passes over the audio stream to map acoustic signatures to potential keyboard characters. During each pass, the model leverages dictionaries to hypothesize character mapping, and check whether the resulting text aligns with standard words. The second model handles the final refinement pass: it ingests thoroughly pre-processed data rather than raw inputs. The method doesn’t stop there: unrecognized keystrokes undergo manual analysis, with analysts injecting educated guesses before re-running the recognition pipeline once again. The goal of looping through these multiple iterations is to achieve complete recognition across the keyboard from an ultra-compact dataset of ideally no more than 200 captured keystrokes. This marks a major shift from legacy methods, which relied on massive training datasets. Clustering the sounds of keystrokes permits grouping similar acoustic profiles together prior to recognition. Notice how distinctly the spacebar sounds stand out: they make subsequent text reconstruction vastly simpler. Source Research results To validate their theoretical model, the researchers created an experimental testing setup: Clustering the sounds of keystrokes permits grouping similar acoustic profiles together prior to recognition. Notice how distinctly the spacebar sounds stand out: they make subsequent text reconstruction vastly simpler. Source The team tested four distinct la… | Kaspersky Security Blog | |||
| –Not rated | – | Cisco Talos Actor usage of AI is exploding. By analyzing artifacts left behind, Talos has created a detailed analysis of how we are seeing adversaries leverage the technology to include development, force multiplication, and vulnerability research. Based on the evidence Talos gathered, guardrails did not provide much protection, with most actors able to convince the models to comply despite the lack of sophisticated techniques or encoding. The pre-existing skill of the actor has a large impact on what they can accomplish with AI. Talos observed novice users able to create malicious capabilities, albeit with limited capabilities and success. Advanced users were able to build astonishing capabilities, pushing the models to create sophisticated and complex outputs. Artificial intelligence (AI) and associated language models are now ubiquitous and heavily used in both personal and professional contexts to streamline tasks and expand capabilities. With AI being used everywhere and by almost everyone, one of the biggest questions is how malicious actors are taking advantage. Fortunately, actors make mistakes and chatbots leave artifacts. Leveraging cloud-based AI models leaves behind a variety of artifacts, most notably a prompt log. These logs can take on a variety of shapes and sizes, but they are left on endpoints that are running various applications, such as Claude Code, CodeX, Cursor, or Gemini. Over the course of our research, we’ve collected a significant corpus of these files and can start discussing the ways we see bad actors leveraging these technologies. In conducting the research, three categories of activity emerged. One was using AI as a malicious software engineer, leveraging AI to write (in some cases) very sophisticated code with clear malicious intentions. Another was actors leveraging AI to scale criminal operations and campaigns. Finally, there were a lot of actors leveraging it for bug bounty or vulnerability research, rapidly accelerating their capabilities of discovery and disclosure. Each category demonstrates how threat actors are currently leveraging AI. Within each category is a wide disparity in sophistication based on the knowledge level of the actors involved. We tried to include use cases to cover the breadth of what we found. Takeaways and high-level findings With the recent disclosures from Hugging Face and OpenAI , it's clear the era of agentic attackers has effectively arrived. In that incident, the models were operating inside a sanctioned evaluation with safeguards deliberately relaxed — but they autonomously escaped their sandbox, found and chained real vulnerabilities, and compromised production infrastructure to reach their objective. The capabilities exist; the only missing ingredient is malicious intent, and it's a matter of time before threat actors supply it. For defenders, this is a wake-up call: Vulnerabilities will surface faster, exploitation will happen sooner, and the actors behind it won't need rest or downtime. As the case studies below show, the central challenge for guardrails right now is supporting legitimate dual-use work — red teaming and vulnerability research — without empowering malicious actors. One of the immediate takeaways is that guardrails are not functioning as expected. We did not encounter any sophisticated encoding or techniques designed to trick the models — most of the time it was a simple “I'm allowed to do this,” and the model complied. When guardrails did engage, they accomplished little. In one instance, we watched an actor abandon a censored model and pivot to an uncensored version, which completed the task without question. In another, a model pushed back on a distributed denial-of-service (DDoS) operator, but by that point the tooling had already been built. This wasn't specific to a single model or platform; it was across the board. The other big takeaway is that an actor's skill level largely determines how effectively AI can be leveraged and how much impact it u… | Cisco Talos | |||
| –Not rated | – | Tenable Research Tenable spent 30 days running frontier AI models against our own code. It didn’t just find bugs — it proved they’re real, with reproducible exploits. That fundamentally changes code security from ranking potential code defects to a much higher signal focused on the findings that matter. Read on to learn how it reshaped our security team's work, what it cost, and why your program is next. Key takeaways: Now code security starts with proof, not suspicions. Frontier AI instantly builds working exploits and proves which flaws are genuinely dangerous in your source code. Now remediations are confirmed issues, not just ranked lists of maybes. The durable asset is the harness, not the model. Frontier AI models get the attention, but the durable asset for security teams is the harness: the orchestration and systems around the model that turn suspected flaws into proven, reproducible exploits engineers can act on. Frontier AI doesn’t replace senior researchers; it makes one as productive as five. The scarce resource is still the expert who writes the threat model and judges what’s real. Buy the compute without funding that person, and you get a very fast way to generate findings no one can use. We’ve been running Claude Mythos Preview against our own code now for over 30 days, and one thing is crystal clear: Code security is fundamentally changing, and we believe there’s no turning back. At Tenable, our security team already had security testing agents that drove our applications, exercised API endpoints, and ran our predefined checks. But until recently, the agents couldn’t handle the harder part of code testing: finding previously unidentified flaws and proving their exploitability. As part of our work testing Anthropic’s Claude Mythos Preview for Project Glasswing , we built an agentic code security harness and powered it with this Anthropic frontier LLM, running it against our code and service repositories with pinned commits to ensure reproducible results. The work of code security is changing, but not in the way hype-driven blogs suggest. And certainly not for free. We’ve found the costs are measured in two currencies: dollars and senior-engineer hours. What follows is an account from the security practitioner’s perspective of Tenable’s internal security team and how we leveraged frontier AI: where the model proved its value, where it fell short, and what you should consider before investing further. From ranking guesses to ranking proof with frontier AI For well over a decade, the scarce resource for security teams was analyst attention. We built a whole discipline around it — reachability heuristics, exploitability guesswork, etc. — all to decide what a human security analyst should look at first in your own code. Now a frontier AI model in a harness collapses that. The ranking doesn’t go away, it gets a proof instead of a guess. When a finding arrives with a working exploit, you’re no longer ranking by how likely it is to matter; you’re ranking by what you’ve already proven does. When an LLM can surface a suspected flaw and drive a working exploit against a running build, the difficult question is no longer “which among the thousands of static findings deserves a human first?” It becomes “which of the code defects are real, and can we prove it?” A bug that shows up with a reproducible proof-of-concept sorts itself. One that can’t be reproduced goes to a validation queue. This does not mean fewer bugs. In fact, it means many more findings, especially early on, because the model surfaces threats and exposures in your source code that traditional tooling never would. Instead, what changes with frontier AI is that practically all of the bugs that reach a human analyst arrive pre-sorted by proof instead of by score. It becomes a short list of things you can already reproduce, plus a holding pen of candidates the harness is still chewing on. The model gets the headlines; the harness does the work This is not to say that the model do… | Tenable Research | |||
| –Not rated | – | Reddit Cybersecurity This report discusses a failure mode in AI language models where the AI, upon correcting a user's technical misconception, proceeds to provide detailed attack tools such as a Kali Linux and Metasploit attack stack. The issue arises because the AI understands the request deeply and supplies precise offensive capabilities without disguise, bypassing current safety measures that focus on detecting disguised or malicious queries. The researchers highlight a gap in safety approaches between the AI's inference of user intent and its actual provision of potentially harmful content. They are working on interventions to address this risk, with full research to be released after coordinated vendor disclosure. | Reddit Cybersecurity | |||
| –Not rated | – | Tenable Research As federal enforcement tightens and states begin stepping in with their own cybersecurity mandates, water and wastewater utilities face a looming wave of hard compliance deadlines, compounded by recent cyber attacks on state water utilities. Key takeaways While the EPA’s national sanitary-survey mandate stalled in court, the agency is aggressively using existing authority, technical guidance, and enforcement alerts to inspect cyber gaps. Community water systems serving 3,301 to 49,999 people, the vast majority of U.S. systems, must certify their Risk and Resilience Assessments (RRAs) by June 30, 2026, under AWIA 2013. New York has already finalized binding cybersecurity regulations for wastewater facilities, setting a regulatory template that other states are expected to follow in 2026 and 2027. Under CIRCIA, utilities will soon be legally required to report significant cyber incidents to CISA within 72 hours and ransom payments within 24 hours. Federal grant programs (SLCGP) and liability protections have been extended through Sept. 30, 2026, but remain tied to unpredictable budget cycles while targeted cyber threats continue to rise. Navigating the new reality of water cyber regulation In 2023, the U.S. EPA made an initial push to fold cybersecurity evaluations into state sanitary surveys. While that effort was stayed in court and subsequently withdrawn, the underlying federal statutory requirements and enforcement drivers remain fully active. Instead of relying on new survey rules, federal and state regulators are actively using existing statutory authority and technical guidance to shift water cybersecurity from voluntary recommendations to enforceable compliance deadlines. The urgency to strengthen cybersecurity for water facilities is underscored by a recent coordinated cyber attack that disrupted water and wastewater utility operations across more than 30 Minnesota communities in late July 2026. Utility cyber regulations and mandates moving forward America’s Water Infrastructure Act (AWIA) 2013 / Safe Drinking Water Act (SDWA) 1433 is still very much in force. Community water systems serving more than 3,300 people are legally required to certify a Risk and Resilience Assessment (RRA) and Emergency Response Plan (ERP) to EPA on a five-year recertification cycle, and that cycle explicitly covers cyber threats, not just physical and natural hazards. Recertification deadlines: Systems serving 100,000-plus people: March 31, 2025 50,000–99,999 tier: Dec. 31, 2025 3,301–49,999 tier, the vast majority of U.S. water systems: June 30, 2026, with ERPs due six months after. The EPA hasn’t stopped pushing on cyber. It’s just doing it through guidance, technical assistance, and enforcement of existing authority rather than new rulemaking. In May 2024, the EPA issued an enforcement alert warning it would step up inspections tied to cybersecurity gaps found in drinking water systems. On Oct. 23, 2025, the EPA released an updated package of cyber tools: Revised Emergency Response Plan guide Cybersecurity Incident Response Plan (CIRP) template Incident-specific checklists Cybersecurity procurement checklist These tools are designed to help utilities fold cybersecurity directly into the RRA/ERP process they’re already required to complete. States are stepping in where EPA stepped back With the EPA’s national sanitary-survey mandate dead, states have started writing their own cybersecurity rules for water systems. New York is the clearest example: In March 2026, the New York State Department of Environmental Conservation finalized amendments to six New York Codes, Rules and Regulations (NYCRR) Parts 616, 650 and 750 , adding binding cybersecurity regulations for wastewater treatment facilities, including mandatory incident reporting and access-control requirements built around EPA’s own cybersecurity guidance, incorporated into the rule by reference. Reporting requirements took effect March 26, 2026. It’s a template other states are watchi… | Tenable Research | |||
| Critical | – | CriticalTenable Research Canada’s new Critical Cyber Systems Protection Act (Bill C-8) introduces a strict 72-hour cyber incident reporting mandate. Find out how Tenable is helping critical national infrastructure operators bridge the IT/OT divide to ensure full compliance. Key takeaways: Bill C-8 introduces stringent new cyber incident reporting requirements and heavy financial penalties for critical infrastructure operators. Eliminating network blind spots with a hybrid IT/OT discovery approach, including Safe Active Querying for isolated, hard-to-reach process-control systems, enables operators to establish a required security baseline. Predictive Vulnerability Priority Rating (VPR) scoring helps you prioritize and focus limited resources on the critical flaws that actually threaten physical safety and uptime. Advanced multi-detection engines and seamless IT workflow integrations accelerate mean-time-to-respond (MTTR) to help both security teams and operators align with a strict 72-hour reporting requirement. With the enactment of Canada’s Critical Cyber Systems Protection Act (CCSPA), commonly known as Bill C-8, the Canadian federal government is laying down a clear framework to protect the cyber-physical systems that are vital to national critical infrastructure security. For designated operators in telecommunications, energy, transportation, and banking, the mandate is clear: Establish formalized cybersecurity programs, mitigate supply chain risks, and — most critically — report cyber incidents to authorities within 72 hours. Failure to comply carries heavy consequences, including penalties that can reach up to $15 million Canadian dollars (CAD). But beyond the threat of fines, Bill C-8 highlights a fundamental operational challenge that many industrial organizations are still struggling to solve: How can you detect, investigate, and report a breach in 72 hours when you lack unified visibility across your converged IT and OT environments? Requirements for meeting Bill C-8's 72-hour incident reporting mandate In modern industrial operations and critical infrastructure, the line between IT and OT continues to blur. The introduction of connectivity (e.g., IoT-connected cameras and building management systems) has optimized processes and service delivery, but it has also introduced new cyber exposures. Today, threat actors do not honor traditional network silos; they frequently compromise a web-facing IT asset or IoT device and move laterally into the operational technology (OT) environment to disrupt physical processes. Meeting a 72-hour incident reporting window is nearly impossible if your security team is relying on fragmented point solutions. Solutions that focus exclusively on passive OT network monitoring often leave massive blind spots — especially considering that IT and IoT devices can constitute up to 50% of an industrial environment. When an incident occurs, teams waste precious hours manually correlating alerts across disconnected tools rather than actively investigating the root cause. To comply with CCSPA and protect uptime, critical national infrastructure (CNI) operators must bridge the IT/OT security divide. Establish your CCSPA cybersecurity baseline The CCSPA requires operators to implement formalized cybersecurity programs. The foundation of any mature security program is a comprehensive asset inventory — you cannot secure what you cannot see. The Tenable One Exposure Management Platform helps organizations eliminate security blind spots by building a complete, unified inventory of all OT, IoT, and IT assets. Tenable goes beyond passive-only network monitoring with our proprietary Safe Active Query technology. This hybrid approach safely communicates with industrial devices in their native protocols to uncover significantly more assets than passive monitoring alone — including dormant process control systems, shadow IT, and unmanaged IoT — without disrupting process integrity or impacting equipment uptime. Prioritize what matter… | Tenable Research | |||
| –Not rated | – | AlienVault OTX General Two newly identified backdoors, OctLurk and SilkLurk, have been targeting government organizations across Central Asia since January 2025. Victims span Afghanistan, Kyrgyzstan, Tajikistan, Uzbekistan, Kazakhstan, and Syria, affecting healthcare, research, government offices, ministries of foreign affairs, logistics, law enforcement, urban planning, and educational institutions. Both backdoors employ heavily obfuscated loaders customized per victim, using machine-specific data for decryption. They deploy multiple plugins for command execution, file manipulation, credential harvesting, keylogging, network scanning, and remote access. The attackers also utilized LurkProxy for network traffic proxying and deployed additional tools including PlugX, Impacket, FSCAN, and Pandora FMS agents. Analysis indicates both backdoors are operated by the same Chinese-speaking threat actor, though attribution to a specific known group remains unconfirmed. The campaigns demonstrate sophisticated persistence mechanisms and ext... | AlienVault OTX General |
Page 1 of 3