artificial intelligence | Breaking Cybersecurity News | The Hacker News

Small artificial intelligence (AI) models may be months away from developing capabilities to self-replicate on IoT devices like smart ovens and off-brand Android TV boxes, according to a warning from Anthropic. The frontier AI lab said small AI models could form autonomous, self-propagating worm botnets that are "challenging to shut down," since "there's no centralized point to shut off." Prior work has demonstrated that AI agents can act like one by scanning their network, executing an exploit, and copying model weights to a target machine across a GPU cluster with an aim to generate tailored attack strategies and use compromised machines as a botnet to run large language models (LLMs) in order to sustain its reasoning and expand in scale. "Such an 'AI worm' could carry its model weights with it as it reproduces (or re-download them from a public hub), and then run inference directly on compromised hardware," Anthropic said . "Eff...

The dangerous part of a stolen identity isn't always the initial login. It's what an attacker can do with that access next. With AI accelerating reconnaissance, exploitation of vulnerabilities, and privilege escalation, attackers can move through systems faster than security teams can investigate. A compromised account that looks legitimate at first could become the starting point for a much larger attack. And that's where traditional identity security faces a serious problem: Checking who gets access isn't enough if you can't control what happens after. See How to Stop AI-Powered Identity Attacks — Register for the Webinar Why Traditional Identity Controls Are Falling Behind

A user opens a browser tab, pastes data into a chat assistant, and nothing is installed. A developer adds a coding assistant as an editor extension, and nothing shows up in the installed programs list. Someone downloads a local model runtime, a program that runs models directly on the machine instead of sending data to an external service, because they were told not to send company data outside. None of this reaches your software inventory. Tasked with controlling AI, an administrator often resorts to blocking well-known AI websites and taking enforcement action without knowing what is actually running. Bitdefender GravityZone AI Visibility and Control (AIVC) is the discovery and risk layer that answers the questions you have, first: Which AI software is present on your protected Windows endpoints, who is running it, and how much risk does each one carry. What Your Software Inventory Was Never Built to Watch AI Visibility and Control covers four classes of exposu...

Some meetings last longer than it took attackers to jump on a newly disclosed security flaw this week. Within hours, they were already testing it. That's not much time to read the warning, let alone do something about it. But speed is only part of the problem. A secure connection didn't always mean a safe destination. People offering to fix a ransomware mess had questions to answer themselves. Malware got better at hiding its tracks. And some systems turned out to be easier to break into than anyone would like to admit. The odd part is how normal much of it looked until someone checked closer. There's more going on here than a few bad bugs and clever attacks. Some of the smaller stories are just as telling. This week's full recap is worth a read, especially the bits that didn't make the big headlines. ⚡ Threat of the Week Atlassian Flaw Exploited in the Wild — Threat actors have begun to exploit a newly disclosed critical security flaw impacting Atlassian...

The personal AI agents rollout is going about as smoothly as expected: Amazon now blocks Meta's Muse because it doesn't identify itself as a bot while browsing; Instinct keeps raising privacy concerns ; and the horror stories from pretty much every personal AI agent keep piling up. Regardless, the tech world is mesmerized by the capabilities and potential of these services. Less than a year after the first release of the open-source agent now called OpenClaw, the juggernauts are putting their weight behind the idea that each of us needs a personal assistant. You know what's coming: your employees will connect their personal agents to work systems. There's no stopping that. Their agents already manage their home time, and managing their work time is the next logical step. For them, that is. Not for your security. Given that personal agents will reach corporate systems whether approved or not, you might be wondering what the best way is to block them. If that’s yo...

Cybersecurity researchers have disclosed details of a two-pronged artificial intelligence (AI)-augmented initial access campaign targeting WordPress sites, e-commerce platforms, and critical infrastructure entities in Ukraine. The activity, which brings together commodity offensive tools like Cobalt Strike with an AI orchestration layer powered by Anthropic's Claude Code, has been attributed to a Russian-speaking, financially motivated initial access broker codenamed CyberXero . The threat actor is assessed to be active since at least July 2026. "The operation runs two pipelines at once: a global, automated campaign against WordPress and e-commerce platforms, and a curated, manual campaign against Ukrainian energy and critical infrastructure," SOCRadar said in a report published last week. The global campaign specifically focuses on known security flaws like wp2shell (CVE-2026-63030 and CVE-2026-60137) and CVE-2026-4815 , an SQL injection vulnerability impactin...

In environments studied for the 2026 State of Agent Security Report , roughly 1,280 third-party products now embed AI. About 282 of them sit behind single sign-on. The other thousand are invisible to identity infrastructure by default, not because anyone hid them, but because an identity stack can only govern what authenticates through it, and most agents never do. That gap is the clearest expression of a shift the security industry is only starting to name. For several years, "AI security" solved a first-party problem: the company decided to use AI, procured licenses, deployed a model behind a gateway, and security pointed controls at the thing the business had chosen. Agents do not arrive that way. They arrive inside software the enterprise already runs, and they arrive without a decision. Why the decision point mattered more than the controls Every control in the first-party toolkit assumes a moment exists: model scanning assumes a model was selected, prompt inspect...

Anthropic on Friday said it's cutting off live internet access for all its internal evaluations until further notice following the discovery of new incidents in which its artificial intelligence (AI) models exhibited misaligned behavior and targeted real websites. The AI company said it identified four broad categories of unintended model actions during evaluations and internal use of Claude - Claude Mythos Preview exploiting SQL or command injection flaws in unspecified third-party software to run commands on a university server, either because its own tools were intentionally limited or because an outside service it needed was unavailable, causing it to use other tools hosted on a third party's site to complete the task. Claude Haiku 4.5 and a non-frontier research model submitting a sensitive form on a real website when it was not authorized to do so. This occurred in scenarios where instructions were ambiguous or due to environment misconfigurations that preve...

Anthropic on Thursday unveiled OSS Scanner as an opt-in vulnerability scanner to help secure the open-source ecosystem using artificial intelligence (AI). "It's an opt-in service informed by our experience using Claude to find vulnerabilities during Project Glasswing," Anthropic said . "Projects that join will receive thorough, periodic security scans by our strongest models at no cost." Anthropic also noted that the outputs of the scanner will be fully model-generated and do not require human review or triage, thereby facilitating faster and more frequent scanning. These reports are expected to be generated by its strongest models, including Claude Mythos. The company pointed out that it expects to use a set of criteria similar to Google's OSS-Fuzz to pick projects, while emphasizing that the process may evolve over time. Project maintainers are advised to provide a short description  explaining the importance of their project in cases where "...

As enterprises race to deploy autonomous AI agents to accelerate business, a new report reveals they are tethered to security architectures built for a different era. The " Horizons of Identity Security " report from SailPoint highlights a critical “velocity paradox,” in which organizations invest in AI-speed business operations while continuing to rely on human-speed security controls, creating a structural failure that legacy approaches cannot solve. The data shows that while businesses have spent years maturing their identity programs for human employees, those same playbooks are fundamentally broken when applied to the ephemeral, autonomous, and rapidly multiplying world of non-human AI agents. A Market Stalled at the Starting Line Despite years of investment in identity and access management, the market's overall security maturity has hit a wall. According to the report, the center of gravity remains firmly planted in the foundational stages, with a combined 6...

The crooks have trust problems of their own. One ransomware affiliate decided to keep the profits for himself. Elsewhere, an attacker left a server exposed, complete with tools and traces of an intrusion. Apparently, keeping things secure is a problem on both sides of the fence. The rest of the week isn't much more reassuring. Malicious code turned up in developer packages and extensions that looked harmless. Familiar online services helped phishing emails appear legitimate. A basic file upload flaw gave attackers a way in, while weak session cookies made impersonation far too easy. Even AI assistants are getting their own instructions hidden inside phishing messages now. What's interesting is the gap between effort and results. Some attacks involve several stages, careful timing, and plenty of tricks. Others get surprisingly far because of a bad design choice or something nobody bothered to check. Both seem to be working well enough. Anyway, here's what else turned u...

The Russia-aligned threat actor known as UAC-0099 has been attributed to a previously undocumented .NET infostealer and remote access trojan (RAT) codenamed ASHVEIN . According to TrendAI, the malware has been put to use in attacks targeting Ukrainian government personnel. The cybersecurity company is tracking the cluster under the name Earth Sirrush (previously SHADOW-EARTH-065). ASHVEIN, which its developers internally refer to as "TelemetryBrowser," brings together credential theft, surveillance, and remote-control capabilities. Its functionality includes credential theft from Chrome and Firefox, GDI-based screenshot capture, file enumeration and retrieval, PowerShell remote shell execution, system fingerprinting, and encrypted command-and-control (C2) communications. "ASHVEIN also hides tasking inside invisible HTML elements," TrendAI said . "Some variants use a GitHub-based dead drop resolver as a fallback mechanism, while delivery methods inclu...

Cybersecurity researchers have disclosed details of a targeted campaign aimed at South Korean financial organizations that used an artificial intelligence (AI) pen testing tool named ARTEX to carry out the attacks. The activity , per CrowdStrike Intelligence, was active from late September to early October 2026, and resulted in data exfiltration from various South Korea-based financial firms , including Shinhan Bank and Yegaram Savings Bank. "In this activity, the threat actor leveraged ARTEX, a recently released open-source agentic penetration testing (pentesting) tool developed in China, alongside large language models (LLMs)," the cybersecurity company said . CrowdStrike said it discovered the campaign after it identified a set of open directories hosted at a Hong Kong-based IP address, exposing Claude Code session histories, Claude memory files, and ARTEX configuration files. The campaign has not been attributed to any known threat actor or group. But evidenc...

The npm package known as " tensorlake ," a TypeScript software development kit (SDK) for Tensorlake applications, sandboxes, and cloud services, was compromised as part of a ChainDrop / Shai-Hulud supply chain attack. The malicious version 0.5.144 "contains obfuscated malware that harvests credentials, exfiltrates secrets, establishes persistence, and executes remotely supplied code," Socket said . Version 0.5.144 is no longer available for download from the npm package registry. An analysis of the compromised release shows that it contains a preinstall hook designed to launch a JavaScript file ("package/lib/setup.mjs"), an obfuscated loader that launches the main credential-stealing and self-propagating worm ("package/lib/Math_Symbol.js") using the Bun runtime. The stealer malware is designed to harvest credentials across local files, CI environments, Kubernetes, and Vault sources. It also drops the HackBrowserData binary, exfiltrate...

A critical vulnerability in LMCache , open-source software that speeds up large language model (LLM) servers such as vLLM, lets an attacker run code on the cache server without logging in, and no fixed version is available. The flaw is in LMCache's  multiprocess mode , where the cache runs as a standalone server that LLM workers reach over the ZeroMQ messaging library. A single network message to that server can run commands as the user the LMCache process runs as. The server can be reached from another machine only when an operator sets it to listen on a routable address, rather than the localhost it uses by default. JFrog disclosed the flaw  on October 7 and assigned it a severity score of 9.8 out of 10, in the critical range, the rating it gives a server bound to a routable address. The vulnerability, tracked as  CVE-2026-105192 , affects LMCache from version 0.3.9, released in October 2025, through 0.5.5, the latest stable release, and is also present in the ...

Cybersecurity researchers are calling attention to a new malware family that has been observed targeting exposed artificial intelligence (AI) and large language model (LLM) infrastructure with an aim to deploy cryptocurrency miners and further expand the scale of the botnet. The financially motivated campaign, dubbed Canto Incognito , has been found to install cryptocurrency miners, including XMRig and Iron, and connects victims to Kryptex, a Russian cryptocurrency mining service. "Compromised hosts are reused to expand the botnet," Lumen Black Lotus Labs said in a report shared with The Hacker News. "Infected servers are turned into scanners and exploit servers, allowing the actor to find and compromise additional vulnerable systems." The malware distributed as part of the campaign has been codenamed PoeLLM owing to what has been described as a "creative" technique that hides the command-and-control (C2) address within a poem the threat actors wr...

The 2026 findings are not just a year-over-year shift. They mark the latest point in a five-year arc where resilience, AI governance, human risk, and board scrutiny are converging inside the systems where work actually happens. For years, the enterprise cybersecurity story has been told as a straight line of escalation: more attacks, more data loss, more pressure, and more urgency. That narrative is still familiar, but comparing the five most recent years of Voice of the CISO research suggests a more useful reading. The CISO role has not simply become harder because every metric is rising at once. It has become harder because the center of risk has shifted and moved closer to the way work now gets done. The latest 2026 findings show signs of progress. Fewer CISOs expect a material cyberattack in the next 12 months, and fewer report material loss of sensitive information than in 2025. But those improvements sit within a longer trend line that is much less settled. Over five years,...

If you’re evaluating an agentic pentesting solution right now, you’ve probably heard the same pitch more than once: point it at a target, and it discovers, validates, and exploits attack paths autonomously, the way a real attacker would. That promise is worth taking seriously. It’s also worth pressure testing, and three questions do the heavy lifting.  What can the assessment actually prove?   When is that proof produced? And,  How much of your environment does the proof cover?  Most evaluations stop at the first step. However, it’s at the second and third ones where validation programs are won or lost. One note on where we stand: Picus builds and sells autonomous pentesting . That is exactly why we can be precise about where it ends, because the limits belong to the method, not to any vendor's implementation, and no roadmap can remove them. The problem in four numbers Four numbers from this year explain why the second and third questions now c...

Anthropic on Tuesday said it's expanding a program that allows vetted cybersecurity professionals to test its advanced artificial intelligence (AI) models with reduced safeguards and blocking classifiers, as the company claimed its Project Glasswing initiative uncovered at least 129,000 verified software vulnerabilities between April and July 2026. The company said it also found an additional 5,500 verified software vulnerabilities between April and October 2026 through open-source scanning efforts. "Of these verified vulnerabilities, more than 33,000 have so far been rated as critical- or high-severity," Anthropic said . "This is likely an undercount, as it is based on survey data from only a subset of Glasswing partners. As such, we expect the true impact to be at least five times higher." The updated program, called the Cyber Verification Program (CVP), features three access tiers, allowing organizations and security teams to apply for one that best a...

Cybersecurity researchers have disclosed details of a "human-operated phishing platform" that impersonates advertising products for artificial intelligence (AI) chatbots like Google Gemini, Anthropic Claude, OpenAI ChatGPT, Perplexity, Meta Muse, and Manus. The products, which claim to offer campaign optimization, spend audits, and business-account connections, are designed with one goal in mind: to capture credentials and multi-factor authentication (MFA) codes via spoofed login windows using the browser-in-the-browser ( BitB ) trick. "Each product was built around the same action: Connect," Island researchers Oleg Zaytsev and Ofek Ronen said in a report shared with The Hacker News. "Clicking it opened a browser drawn inside the real browser. The fake address bar displayed trusted origins such as accounts.google.com or an Okta tenant, while the real browser remained on the phishing domain." "Behind the interface, the platform kept every passw...