Cyber News

Two new unpatched zero-day vulnerabilities in Citrix NetScaler ADC and NetScaler Gateway appliances that allow remote code execution are being actively exploited in the wild, security firm watchTowr said on September 26. Citrix has not confirmed the flaws or published a fix. Some administrators say they have taken appliances offline rather than wait for one to be available. NetScaler ADC and

The ShinyHunters extortion gang is using a URL-encoding trick to bypass web application firewall rules that mitigate the Oracle PeopleSoft CVE-2026-35273 flaw, allowing the threat actors to resume widespread exploitation of a flaw on vulnerable servers. [...]

The Psychedelic Stealer malware distributed via compromised Ukrainian websites using ClickFix-style Cloudflare verification checks is part of a wider malware-as-a-service (MaaS) platform called Lunex. The new findings come from Ontinue, which described the activity as a four-stage attack chain aimed at targeting Ukrainian-speaking users. "The attack chain begins with a fake CAPTCHA page and

Anthropic's Claude Opus 5.5 appears to be changing how it writes, with new analysis showing fewer obvious AI writing patterns, shorter sentences, and simpler wording compared with Opus 5. [...]

Microsoft has paused the rollout of the KB5002907 Microsoft 365 update after users report that it deactivated, or in some cases completely removed, perpetual Office 2016 and Office 2019 installations. [...]

Two third-party GitHub Actions previously compromised in a Mini Shai-Hulud campaign were re-enabled by their maintainer and remained accessible for more than a week despite still pointing to malicious code. [...]

OpenAI says its AI agents uploaded user-provided images to third-party image-hosting services while carrying out research and evaluation tasks. [...]

Google is warning of renewed mass exploitation of a known security vulnerability in Oracle PeopleSoft as part of a campaign targeting multiple sectors globally. The ShinyHunters-linked activity involves the weaponization of CVE-2026-35273 (CVSS score: 9.8), a critical security flaw that could result in unauthenticated remote code execution. The vulnerability was first exploited as a zero-day

The way we talk about AI agents is shifting, and the way we implement them requires an even more fundamental shift. While earlier discourse focused on how quickly organizations could stand up agents and how much productivity they could promise, a string of recent incidents, including a widely discussed intrusion at Hugging Face during an evaluation of OpenAI agents, has spurred organizations to

Details have emerged about a high-severity security flaw in the Elementor Website Builder WordPress plugin that could be exploited by an unauthenticated attacker to create rogue administrator accounts and take control of a site. The cross-site request forgery (CSRF) vulnerability, which has yet to be assigned a CVE identifier, carries a CVSS score of 8.8 out of 10.0. It only affects versions

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) on Friday added two security flaws impacting Microsoft SharePoint and Mikrotik RouterOS to its Known Exploited Vulnerabilities (KEV) catalog, citing evidence of active exploitation. The vulnerabilities in question are as follows - CVE-2026-65660 (CVSS score: 8.8) - A code injection vulnerability in Microsoft Office SharePoint

Kiteworks (formerly Accellion) is urging customers to shut down their systems as a precautionary measure for nine hours over the weekend after it received threat intelligence about an imminent cyber attack. "Kiteworks received credible threat intelligence from federal intelligence authorities indicating that a threat actor may attempt to target some Kiteworks systems," said Frank Balonis, Chief

A U.S. Army soldier who pleaded guilty to hacking into multiple telecommunications companies and stealing mobile call and text metadata for more than 100 million AT&T customers in 2024 was sentenced to 70 months in federal prison today and ordered to pay nearly $300,000 in restitution to victims.

Secure file-sharing software company Kiteworks is urging customers worldwide to temporarily shut down their servers on Saturday for a six-hour window after receiving threat intelligence warning of a potentially imminent cyberattack. [...]

The Clop ransomware gang has moved its data leak site to a new Tor address after confirming its previous server was compromised and defaced through an unpatched Grav CMS flaw that BleepingComputer has learned is an unauthenticated path traversal vulnerability. [...]

When autonomous AI agents "escape the sandbox," the real story isn't rogue machines — it's the same access-control failures we've seen for decades.

A cross-site request forgery (CSRF) vulnerability in the Elementor plugin for WordPress could allow an unauthenticated attacker to create administrator accounts. [...]

In this video conversation, Dark Reading editors discuss some of the news they didn't get a chance to cover, from Google Gemini models breaking containment to ShinyHunters ratting on TeamPCP hackers.

The Cybersecurity and Infrastructure Security Agency (CISA) warns that hackers are exploiting a critical authentication bypass vulnerability (CVE-2026-5430) affecting multiple products from enterprise software provider WSO2. [...]

Anthropic now allows you to run Claude Code via cloud sessions without signing up for the research preview, and it's offering up to $250 in free usage credits, so more users can give it a try. [...]

OpenAI appears to be preparing a new ChatGPT Pro Max subscription that could cost $500 per month, but it's unclear when it'll begin rolling out. [...]

AI agents can operate through human credentials and take actions that existing SOC 2 controls may not distinguish from human activity. Token Security explains why SOC 2 needs to adapt to address the security gaps created by agent identities. [...]

Training human-resource managers in the latest tactics and warning signs goes a long way toward blunting the threat, but automated analysis can help even more.

Two actions-cool GitHub Actions have been disabled for a second time after the repositories became accessible last week, months after they were compromised during the May 2026 Mini Shai-Hulud campaign. The affected GitHub Actions are listed below - actions-cool/issues-helper actions-cool/maintain-one-comment Visiting either of the repositories now shows the message: "Access to this

Cybersecurity researchers have flagged a new version of PamStealer that ensures that the main payload can only be recovered using a server-side decryption chain. The latest artifacts, per Jamf Threat Labs, continue to rely on the same JavaScript for Automation (JXA) dropper mechanism, but modify the lure and the delivery method. "Where earlier variants embedded their payload key material

Microsoft announced it will deprecate the Windows Deployment Services (WDS) server role starting with the next Windows Server release. [...]

A Kosovar national has pleaded guilty to operating Rydox, a large illegal online marketplace that sold stolen personal information, login credentials, credit card details, and cybercrime tools. [...]

Security leaders keep debating whether AI will produce an entirely new class of cyberattack. The nearer change is quieter and already visible: AI has made a failed attack cheap to retry. The routine version looks like this. An attacker lands on a low-privilege cloud account, and the first try at privilege escalation goes nowhere. That dead end used to cost hours of documentation reading,

Cloud Range, vendor of cyber range platform for validating human and AI readiness, announced official launch of its... The post Cloud Range introduces AI Readiness Framework to validate AI agents before operational deployment appeared first on Industrial Cyber.

Optigo Networks announced the public launch of the Optigo AI Assistant, a purpose-built assistant embedded in Optigo Visual... The post Optigo Networks launches AI Assistant to address building automation network expertise gap appeared first on Industrial Cyber.

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) published on Wednesday a framework for establishing the ‘Quality Era’... The post CISA launches CVE Quality Era framework to improve vulnerability data quality, infrastructure, program governance appeared first on Industrial Cyber.

New data from the Ridge Security Research Team disclosed CVE-2026-42542, a pre-authentication vulnerability in TDengine, an open-source distributed... The post Ridge Security warns of high-severity TDengine vulnerability that can disrupt industrial telemetry and monitoring appeared first on Industrial Cyber.

Cryptocurrency exchange Bitget said suspected North Korean threat actors have stolen $351.6 million from its hot and warm wallets. "At 18:31 UTC on September 24, 2026, Bitget's security systems identified unauthorized transfers involving a limited number of hot wallets," BitGet said in a post shared on X. "Bitget's cold wallets and the overwhelming majority of platform assets remain

Microsoft has confirmed that some users may experience desktop loading issues, including black screens, after installing the August 2026 preview updates and subsequent updates. [...]

The Canadian Centre for Cyber Security has warned that a now-patched Roundcube Webmail vulnerability is being actively exploited in the wild. The vulnerability in question is CVE-2026-48842 (CVSS score: 8.1), a pre-authentication SQL injection in the virtuser_query plugin of Roundcube Webmail versions 1.6.x before 1.6.16 and 1.7.x before 1.7.1. The issue stems from a preg_replace() backslash

​Cryptocurrency exchange Bitget disclosed today that suspected North Korean hackers have stolen $351.6 million from its hot and warm wallets. [...]

A storm is raging in the form of cyber sabotage, disinformation, and drone attacks on European nations, particularly those that provide material support to Ukraine.

Introduction

A flaw in Cloudflare Containers let a paying customer read data that other customers' containers had left behind on the same server, Cloudflare and the researchers who found it said on Thursday. The data came from disk space that earlier containers had used and given up, not from any live workload, and an attacker could not choose whose data they got, according to Cloudflare. The company

The U.S. Cybersecurity and Infrastructure Security Agency (CISA), on Thursday, added two critical security flaws impacting WSO2 and Adobe Commerce and Magento to its Known Exploited Vulnerabilities (KEV) catalog, based on evidence of active exploitation. The vulnerabilities are listed below - CVE-2026-5430 (CVS score: 9.8) - A path traversal vulnerability in WSO2 API Control Plane,

arXiv:2609.29173v1 Announce Type: cross Abstract: Video see-through extended reality (VST XR) systems commonly use headset screenshots or captured frames as proxies for the user's first-person visual context. However, the system-captured view and the user's effective visible field do not necessarily coincide: a screenshot records a rectangular machine-readable frame, whereas the user's effective visible region can be more constrained and non-rectangular. This paper studies this human-system view mismatch in VST XR. We formalize the relationship between the system-captured region and the human-visible region by defining their co-visible, system-only, and human-only regions. \rev{We then conduct a pilot-level boundary measurement on Meta Quest 3, revealing a clear mismatch between the rectangular screenshot frame and the approximate human-visible boundary. Building on this model, we analyze how view mismatch can affect screenshot-based XR sensing and downstream vision-language model tasks. Through four representative case studies, we illustrate potential risks and failure modes including prompt injection, privacy leakage, human-invisible information bias, and missing human-visible information. Our results show that view mismatch is not only a geometric artifact, but can also introduce security, privacy, and reliability concerns for AI-integrated VST XR systems.

arXiv:2609.29757v1 Announce Type: new Abstract: Publicly exposed large language model (LLM) infrastructure creates a growing attack surface, yet real-world targeting remains poorly understood. We present Ollure, a low- and medium-interaction honeypot that emulates the Ollama API without a backend LLM. Spanning four deployments across cloud and university networks, Ollure operated for 84 days and recorded 290,887 interactions from 2,793 unique source IP addresses. Most of the activity consisted of automated discovery, fingerprinting, and model enumeration. However, we also observed concrete exploitation attempts against both the infrastructure and LLM layers. These included model management abuse, path traversal and SSRF probes, RCE and cryptocurrency mining payloads, resource exhaustion attempts, prompt injection, information extraction, and agent-oriented tool use. Our results provide empirical insight into real-world threats against exposed, self-hosted LLM services.

arXiv:2609.28613v1 Announce Type: new Abstract: Most studies of prompt injection focus on generative agents, leaving their effects on models with schema-defined outputs unclear. We examine these effects in Jev, a non-generative decision model, using 510 reconstructed InjecAgent cases. Malicious content shifts action probabilities but rarely causes Jev to select the attacker's target. Override markers reduce this influence, while claims of contextual relatedness have small effects. Adaptive attacks using score feedback double the mean highest attacker-target probability found during optimization, while success on fresh validation calls rises from 1.8% to 3.5%. Exploratory analysis links these successes to small initial decision margins or greater attacker control over the observation. Together, these findings show that schema-defined outputs change but do not eliminate prompt-injection risk, highlighting the need to evaluate how untrusted content influences choices within the allowed action set.

arXiv:2609.29287v1 Announce Type: cross Abstract: Large audio-language models (LALMs) expand language models to process and interpret audio, but also expose them to heterogeneous audio jailbreaks. We ask whether successful jailbreaks reflect failures to recognize harmful intent or failures occurring after such recognition. Layer-wise probing reveals the latter: risk-related information remains decodable from intermediate representations, yet the internal risk signal fails to translate into refusal in later-layer processing. We identify this discrepancy as the risk-to-refusal gap. Building on this finding, we propose AEGIS, a detect-then-intervene defense whose mid-layer risk gate selectively activates downstream safety adapters. Across six LALMs and three heterogeneous audio jailbreak benchmarks, AEGIS reduces the average unsafe rate from 17.9% to 0.4%, while causing only a marginal increase in over-refusal on benign inputs. These results establish selective internal intervention as an effective path toward more robust refusal in LALMs. The code is available at https://github.com/azzzzliao/aegis-audio-defense.

arXiv:2609.26176v2 Announce Type: replace Abstract: Encoded-prompt attacks are evaluated almost entirely on their harmful arm: a benchmark sends obfuscated harmful requests and reports how often the model complied. A high refusal rate there is reported as safety, and it is equally consistent with a model that has stopped telling the request apart from anything else in the same format. We run the benign arm through the same transformation, and the two cases are far apart. Across four 7-8B models spanning three base families and four post-training recipes, refusal of harmful homoglyph-encoded prompts spans 0.08 while the same four span 0.57 on the identical requests in plaintext. What the encoding destroys is not refusal but the harm gap: on one model the gap between harmful and benign refusal falls from +0.82 in plaintext to exactly 0.00 under the encoding, and a benchmark reading only the harmful arm scores that model and one retaining a +0.61 gap identically. Running the cell such benchmarks leave out (plaintext content wearing the attack template, with nothing obfuscated) shows that on two of the four models the loss is caused by the protocol rather than by the character transformation, and on a third by the characters. Across a full SFT -> DPO -> RLVR pipeline the harm gap rises by +0.26 with a paired interval excluding zero while the standard harmful-arm metric registers no resolved change at all. We report twelve instrument defects, each with the control that caught it, including a binary jailbreak judge that fires on 0.61-0.70 of responses to plaintext benign prompts; six of the twelve inflate apparent safety, which is the direction a broken safety evaluation fails in by default.

arXiv:2609.28564v1 Announce Type: new Abstract: Agentic video-generation systems close a loop between a generator and a verifier: an LLM plans shots, calls a text-to-video model, and a multimodal judge decides whether the result satisfies the request. To diagnose where a long workflow fails, recent harnesses deliberately show the judge more than the video-the agent's execution trace, its plan, the narration it synthesized. We ask whether this auxiliary text moves the judge's verdict on purely \emph{visual} requirements, holding the frames fixed. On a benchmark of 109 generated two-event clips with manual labels, in which the requested event is either visibly completed or visibly missing, a trace that reports a successful tool call makes three open-weight Qwen-VL judges (7B, 8B, 32B) accept $78$--$90\%$ of the failures, up from $7$--$19\%$ without text, and a contradicting trace makes them reject up to $100\%$ of correct clips; an instruction to ``use only the frames'' does not remove the effect. Frontier closed judges are essentially unmoved on the same clips, showing that the vulnerability is a property of the judge's learned trust in tool logs rather than of the task. Plan-derived text carries no clip-specific information, so it can only shift a judge's operating point, and in a repair loop that shift becomes a cap on the true pass rate that no repair policy can exceed; the cap matches simulation to two decimals. In the loop, contamination is exploited without any adversarial agent: an honest LLM planner that always regenerates ends with a judge pass rate of $1.00$ and a human-labelled pass rate of $0.28$, and a pipeline in which a cheap checker writes its verdict into the trace launders that checker's errors into a stronger final judge ($0.69$ false accepts).

arXiv:2609.29647v1 Announce Type: new Abstract: Modern AI agents routinely cross trust boundaries: they ingest untrusted content, combine it with privileged instructions, persist intermediate beliefs in long-term memory, and invoke privileged tools. This creates an attack surface in which malicious payloads can enter through model inputs and cause harmful tool actions. Yet current governance stacks remain application-level middleware that share a process trust boundary with the agents they monitor. We argue that agents need an operating-system substrate providing mandatory, non-bypassable services for identity, input mediation, memory governance, and execution control. We introduce AgentKernel, a trust-native agent operating system built around the premise that security must be a first-class design constraint. AgentKernel wraps the agent lifecycle in a mandatory enforcement boundary organized into four pillars: Identity, Perception, Cognition, and Execution. Each pillar adapts classical OS security principles to failures at the semantic plane, including delegation abuse, prompt injection, memory poisoning, and tool misuse. AgentKernel treats structural security as a capability multiplier. Kernel-managed identity supports trustworthy cross-organization collaboration; graduated perception replaces brittle single-point filters; information-flow-controlled memory improves retrieval fidelity while limiting poisoning; and semantic-to-kernel enforcement permits broader tool privileges behind a non-bypassable boundary. We position AgentKernel as the missing OS layer beneath orchestration frameworks, agent runtimes, governance platforms, and execution sandboxes, and use systematic comparison and security analysis to show how a single integrated architecture can enforce security across the full agent lifecycle.

arXiv:2609.30217v1 Announce Type: new Abstract: A central concern in AI safety is that agents may treat oversight as an obstacle when it conflicts with completing their goals. We study instrumental evasion, the propensity of LLM agents to circumvent runtime monitoring as a means of completing ordinary tasks. We introduce EvasionBench, a benchmark of 50 diverse task-policy pairs in which completing the task requires an operation prohibited by a runtime monitor. Agents know that their tool calls are monitored and are prompted to continue working when they pause. Across our evaluations, best-of-3 evasion attempt rates reach up to 98% and success rates up to 88%, with substantial variance across models. Claude Fable 5.1 succeeds less often, but frequently makes creative attempts to circumvent the monitor. Evasion generally increases with test-time compute, with higher evasion rates at greater reasoning effort and token use. Traces show that agents encode prohibited commands, decompose operations across tool calls, and retry until relevant context leaves the monitor's history. We also find that GPT-6 Astra's low evasion rate comes with overrefusal, as it frequently abandons otherwise solvable tasks under a denial-of-service prompt injection. Our findings show that ordinary task pressure can lead to adaptive attempts to evade runtime monitors without an explicit adversarial objective. Effective oversight must therefore remain robust against repeated attempts, as the persistence that helps agents solve difficult tasks can also drive them to circumvent their guardrails.

arXiv:2609.29429v1 Announce Type: cross Abstract: Detectors of alignment failures screen deployed language models and score alignment benchmarks. Most are generative judges that spend a decoding pass on every criterion, and classifiers that read token probabilities, such as Llama Guard, still score one fixed label per call. Jev, a model trained with reinforcement learning for calibrated decisions (RLCD), answers many typed questions about one input with calibrated probabilities in a single call. Whether it detects alignment failures has not been measured. We present RLCDAlignBench, which benchmarks Jev on ten alignment failures: sycophancy, jailbreaks, deception, prompt injection, hallucination, privacy violation, social bias, reward hacking, concealing uncertainty, and power seeking. It spans 44 benchmarks and five target models, labelled by each benchmark's scorer and, on two, by humans. Many of these failures are relational, defined against a reference, such as the user's belief or an injected instruction, that the response alone does not reveal. Our key idea is therefore to vary what Jev is asked separately from what it sees: the question's wording and answer type on one side, the fields of the input on the other. A single generic question reaches a median AUROC of 0.886 zero-shot and beats supervised baselines on most benchmarks. Question wording matters little, while context matters more, mostly through fields that encode the label. Jev matches the reference scorer's agreement with human labels, surfaces label defects in existing benchmarks, and costs 63x less than LLM-judge scorers. Code and data: https://github.com/sumleo/RLCDAlignBench.