artificial intelligence | Breaking Cybersecurity News | The Hacker News

Google's Gemini model has become the latest artificial intelligence (AI) system to access the internet and break into other companies during a cybersecurity evaluation. The development was first reported by The Wall Street Journal. The incidents occurred in May 2026 as part of a test run conducted by Israeli company Irregular. The evaluation partner was also involved in similar hacks disclosed by OpenAI, Anthropic, and Meta. According to the Journal, the model gained access to a protected system after repeatedly guessing its password. Two other cases related to the model finding credentials in a public repository, allowing it to obtain unauthorized access to protected systems. However, unlike other incidents observed in the case of Anthropic and OpenAI, the Gemini model ended the intrusion after finding that it had breached a real company's system. Irregular is said to have notified Google of the incidents in July 2026. In a report published last month, Irregular pinn...

A security researcher has released working exploit code for four Linux kernel flaws that each let a local user gain root, the highest level of access on a machine. Kernel maintainers have fixed all four over the past few weeks, so a system running an up-to-date kernel is not affected. But the exploit code is now public, and any machine still running an older kernel should be updated. The flaws are called DirtyAH6 , TUNderflow , PPPoEject , and DiagSpill . Researcher Asim Manizada found them and reported them to the Linux kernel security team in mid-July. He published a technical  write-up  with working exploits on September 18, after a coordinated hold with Linux distributions so the fixes could be released first. So far, there are no reports of the four being used in real-world attacks. The exploits are Manizada's own, are tuned to specific kernel builds, and can crash a machine, so they are meant for isolated test systems. Still, public exploit code raises the ris...

Microsoft has released fixes for a maximum-severity security flaw in Azure AI Foundry that could be exploited to achieve privilege escalation. No customer action is required. The vulnerability, tracked as CVE-2026-85889 , carries a CVSS score of 10.0. "Missing authentication for critical function in Azure AI Foundry allows an unauthorized attacker to elevate privileges over a network," Microsoft said in a Thursday advisory. Azure AI Foundry, also called Microsoft Foundry, is an enterprise platform designed to build, deploy, and manage generative artificial intelligence (AI) applications and agents. The Windows maker credited security researcher Rémy Marot (@R_Marot) for discovering and reporting the flaw. There is no evidence that the issue has been exploited in the wild. Also patched by Microsoft in recent days are a number of other critical flaws -

A flaw in four widely used AI coding agents lets someone who controls a plugin's code repository swap the plugin an agent installs for a malicious one, even when the agent locked that plugin to a specific reviewed version, security firm  Air Security said on Thursday . The firm said Anthropic has patched the flaw in Claude Code 2.1.179 and OpenAI in Codex 0.146.0, that GitHub Copilot has no fix, and that Google will not patch the Gemini CLI, which it is retiring. The agents install add-ons called plugins from online marketplaces. To stay safe, a marketplace locks each plugin to a single reviewed version by its commit hash, a long string that identifies an exact snapshot of the code. Air found that the agents fetch that snapshot but never check that the code they end up with actually matches it. A branch is a named line of code in a repository. On a code host that lets someone create a branch whose name is made to look like the commit hash, the owner of a plugin's reposit...

A financially motivated threat actor has been linked to the development and distribution of a JavaScript (JS)-based information stealer known as PhantomRaven via the npm package registry. "The developer likely wrote the malware using a large language model (LLM), an assessment made with high confidence based on verbose comments, placeholder code, and statistical token-analysis patterns," CrowdStrike's Counter Adversary Operations said in an analysis published this week. PhantomRaven was first flagged by Koi Security and DCODX in late October 2025, calling attention to a slopsquatting and typosquatted campaign in which more than 100 malicious packages were uploaded to npm to steal authentication tokens, CI/CD secrets, and GitHub credentials from developers' machines. The software supply chain attack used these packages as a cover to retrieve a remote dynamic dependency (RDD) from an external server so that the libraries themselves are not flagged by security ...

Cybersecurity researchers have flagged a new Android malware called RatHat that's assessed to be operated by China-based threat actors and features an artificial intelligence (AI)-powered system to navigate and control compromised devices. "Distributed primarily via targeted smishing (SMS/text phishing) and malvertising campaigns leading to deceptive third-party download portals, RatHat uses an automated multi-stage infection pipeline," Zimperium researchers Gianluca Braga, Vishnu Pratapagiri, and Fernando Ortega said . "Once deployed, it pairs Accessibility abuse with autonomous local ADB (Android Debug Bridge) self-pairing to break out of the standard Android application sandbox, staging independent native daemons that execute with shell-level privileges." RatHat is propagated via deceptive phishing sites promoted via malvertising, smishing campaigns, and third-party forums that trick unsuspecting users into installing malware-laced APKs. These pa...

Attackers keep finding new keys. The funny part is that defenders keep inventing where to store them. This week, those keys sit in AI tools, exposed services, old bugs, weak logins, and software sold like a monthly subscription. Some attacks use new tricks. Others just reuse what was already lying around. Both work often enough. So the threat landscape is not getting cleaner. It is just getting more places to make the same mistake. Here’s what showed up this week. The threats change every week. Subscribe, and we’ll alert you when each new ThreatsDay Bulletin is out.

Malicious code running inside a Docker Sandboxes virtual machine on macOS could escape the project directory shared into it and read or change files anywhere else on the host, Docker warns in a  security announcement  on September 15. The escape runs with the rights of the host account that runs the virtual machine. The flaw,  CVE-2026-77179 , is rated Critical, affects versions 0.28.0 up to but not including 0.42.0 on macOS, and was fixed in  0.42.0  on September 7. Docker Sandboxes runs each AI coding agent in its own small virtual machine with the project directory shared in. The code that could escape is whatever runs inside that machine, such as a coding agent that has been turned against its user, or anything malicious the agent installs and runs. Docker has not reported any exploitation. CISA's added assessment on the CVE record lists exploitation as none, and the flaw is not in CISA's Known Exploited Vulnerabilities catalog as of the ca...

A new CVE drops. Your scanner finds it. The severity score looks ugly. But that still does not answer the question that matters: Can it actually be exploited in your environment? Mythos-class AI is compressing the time between disclosure and working exploitation, while many security programs still validate risk on weekly or quarterly cycles. The dangerous gap is no longer just technical. It is measured in time. 📅 Save Your Spot Today: How to Prove You're Ready for Mythos-Class Attacks . Can’t join live? Register anyway. We’ll send the webinar recording after the session, so you can watch it later, whenever it fits your schedule, and still see the full, fresh CVE-to-validation workflow. Stop prioritizing on severity alone A high score tells you a vulnerability could be serious. It does not prove that an attacker can use it against you. Security teams need faster answers: Is the affected asset exposed? What attack techniques does exploitation require? Do ...

Attackers now weaponize new vulnerabilities in about five days (Mandiant, part of Google Cloud). The median organization takes 43 days to patch one (Verizon DBIR 2026). A new free guide explains how autonomous AI agents are closing that gap, and what security leaders must demand before pointing one at production. TL;DR Exploitation is now the front door. It starts 31% of breaches (Verizon DBIR 2026), the #1 initial-access vector, while annual pentesting leaves an estimated 90% of the estate untested. The capability is proven, not projected. An autonomous system topped HackerOne's US leaderboard in 2025 (XBOW), and peer-reviewed agents exploited 87% of one-day flaws unaided (Fang et al., 2024). Continuous beats periodic, measurably. Programmatic testing makes teams 4.5x more likely to fix criticals within three days (Cobalt, 2026). It's an AI agent in your production. Hold it to a bar. Provable coverage, an independent validator, blast-radius guardrails, and a...

OpenAI on Wednesday disclosed six new instances of "unexpected or concerning model behavior" that took place over the past six months, while sharing a new framework for reporting, tracking, investigating, and disclosing model misalignment in a bid to improve transparency. "As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research," OpenAI said. "We do not believe that the AI industry has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer." "Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves." The six incidents are outside of the recently disclosed misaligned activity targeting Hugging Face , DseWiki , and RubyGems . Details of the inc...

Security researchers at Forever Security have shown that one ordinary browser extension could take control of the AI assistants built into five Chromium-based products: Gemini Live in Chrome, Perplexity Comet, Microsoft Edge, Opera Neon and the Claude in Chrome extension. Once the extension was installed, it could access each product's built-in AI with a single click. On Comet, Edge, Opera Neon, and Claude in Chrome, it could drive the AI agent to act on behalf of the attacker; on Chrome and Comet, it could read files from the user's computer, and on Chrome, it could also switch on the camera and microphone. The findings are researcher demonstrations, not attacks seen in the wild, and each requires the attacker's extension to be already running in the victim's browser. These products all work the same way. The AI has a "body" inside the browser that can see the screen, open files, use the camera, and take actions, and a "brain" that runs on th...

Mandiant says an attacker hijacked an active AI coding-assistant session at an unnamed software-as-a-service provider and later spread  Shai-Hulud  across about 100 internal code repositories. Before the repository spread, the assistant recommended software that the attacker had poisoned, and the recommendation was accepted. The worm stole repository secrets and source code for the company's products. The case appears in  Mandiant's September 2026 report . The public case study does not say when the intrusion happened or how the attacker took over the active coding-assistant session. How the Attack Unfolded After the recommendation was accepted, the attacker used the developer's active session to install an infostealer through a poisoned PyPI package. The attacker also stole GitHub OAuth tokens. The attacker then deployed the self-spreading Shai-Hulud worm across approximately 100 internal code repositories. The attacker also poisoned a package in the company...

Introduction Security teams have gotten pretty good at testing against what can hurt them. Can this EDR agent catch this payload? Will my organization fail the phishing simulation? Does this SIEM rule fire on this particular technique? And, in more mature organizations, this testing happens continuously rather than as a one-off exercise. But no matter how much you validate against these exposures, it doesn't fix the main problem the industry is facing: these are isolated, disconnected testing.  And real attackers, increasingly AI Powered ones, don't test techniques one at a time. They chain them. A phishing email leads to a credential harvest. That harvest leads to an initial foothold. The foothold leads to privilege escalation, then lateral movement, then data staging, then exfiltration… until the damage is irreversibly done.  Any one of those individual steps might be something a security control is theoretically capable of catching - but there's just too many po...

WordPress has announced it's launching an automated security review for every release of a plugin before it's distributed through the WordPress.org update API so as to analyze it for potential security issues and ensure there are no risks involved. "New plugins are reviewed before they enter the directory, but updates ship continuously after that," David Perez, WordPress Official Plugin Repository Team Co-Lead, said . "A plugin can be secure today and introduce a vulnerability, or malicious code, in a future release." WordPress said the lack of a "consistent review step" between the commit of a release and the release of a plugin to downstream users meant that it could open the door for malicious attacks. The content management system (CMS) platform noted that its automated review detected a backdoor committed to a release of a plugin with about 20,000 active installations on July 28, 2026. Because the release was within a cooldown window, ...

AI keeps showing up in the wrong places. Attackers are using it to speed up exploits, test defenses, and automate more of the job. Some models are also crossing lines on their own. That is not a great combination. The rest of the week is more familiar: old bugs still working, fresh exploit chains, exposed systems, weak defaults, and simple paths that should have been harder to abuse. A few of these stories are clever. Most are just easy. Here’s what mattered this week. ⚡ Threat of the Week OpenAI Agents Behind May 2026 Attack on RubyGems — The "major malicious attack" that targeted RubyGems in May 2026 was the work of a swarm of OpenAI agents, according to researchers. The event was driven by a cluster of OpenAI agents that engaged in en masse publication of thousands of packages to RubyGems in May and June 2026. "The swarm behaves extremely similarly to the German-wiki agents we previously found," researchers Spencer Kitts, Thomas Larsen, and Sydney Von...

There's a lot of noise around AI and cybersecurity right now. What’s actually important is far simpler, if often lost in the hubbub. Vulnerability discovery is getting faster and happening at a much greater scale, while defenders still have to work out which findings actually deserve their action. In the first half of 2026, a whopping 35,853 CVEs were published, roughly 49% more than in the year before. Yet only 495 were catalogued as exploited in the wild during that same period, and 116 were already under attack on the day they became public. Meanwhile, Anthropic’s own disclosure data shows Mythos-class models surfacing 26,153 vulnerability candidates in open-source software, with only 421 of those getting patched upstream. That small exploited subset is a very important point. It tells defenders that treating every vulnerability with a High or Critical CVSS rating as an emergency is not only impossible, it’s actually the wrong model . The critical task security teams fa...

Microsoft has disclosed details of two campaigns in which threat actors are abusing third-party email delivery infrastructure to blast financial fraud scam messages and using passkey-themed social engineering to breach cloud environments. The first campaign, per the tech giant, involved sending over a million scam emails between August 3 and 5, 2026, by masquerading as chief executive officers (CEOs) of various target companies, aiming to persuade accounts payable departments at those firms to initiate Automated Clearing House ( ACH ) transfers for a supposed ServiceNow annual subscription. Evidence indicates that the operators behind the campaign have leveraged generative artificial intelligence (AI) to facilitate the creation of email templates and draft emails tailored to their recipients. The activity primarily singled out enterprise users in the U.S., spanning IT services, consumer goods, real estate, and discrete manufacturing sectors. "The campaign follows steps bef...

Over the past year, we watched a new class of alert appear in enterprise security operations centers and grow faster than anything else in the stream: alerts that were triggered by AI tools and agents. Not attacks against AI, but the ordinary, everyday footprint of an organization using it, from developers running coding agents and non-technical staff signing consumer AI tools into corporate accounts. We reviewed AI-related activity across numerous enterprise environments. Two numbers frame everything that follows. AI-related alerts still account for only 0.43% of all SOC alerts. And that share is climbing every single month, up 685% between February and June 2026. AI is a small slice of the alert stream today and the fastest-growing slice at the same time. What makes those alerts worth a security team’s attention is not their volume but their composition. We sort everything an AI agent triggers in a SOC into three buckets: real attacks, risks, and noise, with the split being 94....

The "major malicious attack" that targeted RubyGems in May 2026 was the work of a swarm of OpenAI agents, according to a new report published by researchers Spencer Kitts, Thomas Larsen, and Sydney Von Arx. On May 12, Maciej Mensfeld, senior product manager for software supply chain security at Mend.io, disclosed details of a coordinated cyber attack that targeted the package manager for the Ruby programming language with hundreds of junk gems, prompting the maintainers to suspend new user sign-ups for about four days. In a follow-up analysis, Socket highlighted a campaign dubbed GemStuffer that involved a cluster of more than 150 gems that used the package registry as a data exfiltration channel and staged public data scraped from U.K. local government democratic services portals. At that time, the software supply chain security company noted the activity shares the "same abuse pattern" as the broader RubyGems spam-publishing incident. "It's not...