Cyber security
8 items · 50 articles in this topic
Edge and management systems dominate: vCenter, BIG-IP, Cisco Expressway and SonicWall. A second cluster is third-party and supply-chain incidents behind bank and SaaS outages. Several breach disclosures and an AI-assisted attack on the European Commission round out the day.
- 1
Cisco Expressway CVE-2026-76466 is exploited, and F5 BIG-IP has five new critical flaws, as covered above. Treat both as edge-exposure priorities alongside the vCenter fix.
- 2
SonicWall says Gen 6 firewalls will no longer receive security patches, even for critical flaws, and urges immediate replacement. Any Gen 6 unit still in service is now permanently exposed to new bugs.
- 3
N-able says attackers who breached a server caused a worldwide outage of its Mail Assure and SpamExperts Hosted Cloud services. A compromised email-security vendor affects every customer downstream.
- 4
OpenSSL disclosed a buffer overflow in EVP_EncodeBlock(), CVE-2026-121481, rated CVSS 9.8. The summary gives no affected versions, so check the advisory. Separately, Linux KVM CVE-2026-99312 lets guest VMs read host memory on AMD hosts with nested virtualization and is fixed in shipped kernels.
- 5
ASUS fixed three critical flaws in Control Center Express v1.7.29, including a hard-coded key that allows remote admin control. Synology published advisory SA-26-21, with details on its site. Windows also has an unpatched zero-day, FastUnlock, which lets someone with physical access run programs on a locked PC with limited privileges.
- 6
The European Commission is investigating an AI-assisted cyberattack that took HR and medical reimbursement systems offline and may have exposed personal data. Equinor is recovering from a cyberattack with no timeline, though oil and gas production is unaffected.
- 7
Breach disclosures: SBC Medical Group told the SEC two incidents exposed data of 1.6 million customers and employees. Transport for London may have exposed account details of about 265,000 Oyster photocard users.
- 8
Debian's APT is disabling and removing older complex features to shrink its attack surface after a MITM vulnerability, a move tied to advances in AI security discovery. The ROMULUS ClickFix platform is delivering a previously undocumented RAT similar to CNCMachineRMS, alongside Legion Loader.
HighPhoronix ↗
More in Cyber security · 42 articles›
F5 urges customers to urgently patch five critical BIG-IP flaws enabling denial-of-service or remote code execution.
BleepingComputercvepatchf5rceFeed excerpt
US cybersecurity company F5 warned customers to patch five critical BIG-IP vulnerabilities that attackers can exploit to trigger denial-of-service (DoS) or gain remote code execution. [...]
China-nexus group Armory Castle exploited a zero-day, CVE-2026-123890 (CVSS 9.8), in VMware vCenter Server authentication to steal data.
The Hacker Newscvezero-dayvmwareespionageFeed excerpt
The suspected China-nexus cyber espionage group known as Armory Castle (formerly UNC5221) has been attributed to the zero-day exploitation of a critical security flaw in VMware vCenter Server in early September 2026. The vulnerability, tracked as CVE-2026-123890 (CVSS score: 9.8), is a case of authentication bypass in the VMware Directory Service that could be exploited to gain unauthorized
Broadcom patched a critical VMware vCenter Server authentication bypass that Chinese-linked hackers exploited as a zero-day since September.
BleepingComputercvezero-dayvmwarechinaFeed excerpt
Broadcom has released security updates to fix a critical VMware vCenter Server authentication bypass vulnerability that Chinese-linked hackers have exploited in zero-day attacks since September 2026. [...]
Broadcom published an advisory for a critical VMware vCenter flaw (CVSS 9.8) allowing network attackers to bypass authentication, with no workaround.
The Hacker Newsvmwarecveauthentication-bypasspatchFeed excerpt
A critical flaw in VMware vCenter lets an attacker with network access bypass authentication and gain unauthorized access to the system, Broadcom, which owns VMware, said in an advisory published today. Fixed versions are out for all supported branches, and the company lists no workaround. vCenter is the central console for managing an organization's ESX hosts and virtual machines. The
The PHP Foundation released security updates fixing seven flaws, including CVE-2026-115560 (CVSS 9.2) enabling remote code execution.
The Hacker NewscvephprcepatchFeed excerpt
The PHP Foundation has released security updates to address multiple critical security flaws that could be exploited to achieve remote code execution on affected deployments, including CVE-2026-115560. The list of vulnerabilities is as follows - CVE-2026-115560 (CVSS score: 9.2) - An improper input validation vulnerability that could result in remote code execution, file modification,
CVE-2026-99312 in Linux KVM lets guest VMs read host memory on AMD hosts with nested virtualization, fixed in shipped kernels.
The Hacker NewscvelinuxkvmvirtualizationFeed excerpt
A flaw in KVM, the hypervisor built into the Linux kernel, let researchers read about 127 GB of a cloud server's memory from inside a virtual machine of their own on DigitalOcean. On a shared server, that memory can include other customers' virtual machines. The flaw, tracked as CVE-2026-99312, affects only hosts with AMD processors and nested virtualization enabled. Fixed kernels shipped
BBC reports a secret nationwide UK operation secured internet-connected fridges that hackers had compromised.
Hacker News (front page)iotbotnetukFeed excerpt
Article URL: https://www.bbc.co.uk/news/articles/c6eq49zyx7p4o Comments URL: https://news.ycombinator.com/item?id=50572430 Points: 52 # Comments: 37
The ROMULUS ClickFix platform is delivering a previously undocumented RAT similar to CNCMachineRMS, alongside Legion Loader.
The Hacker NewsmalwareclickfixratFeed excerpt
Threat actors have been observed using the ROMULUS ClickFix malware delivery platform to deliver a previously undocumented remote access trojan (RAT) that shares similarities with CNCMachineRMS. "ROMULUS likely operates as a delivery service, distributing CNCMachineRMS and Legion Loader through separate chains while retaining ClickFix and signed binary sideloading techniques," Blackpoint
A researcher disclosed an unpatched Windows zero-day, FastUnlock, letting someone with physical access run programs on a locked PC with limited privileges.
The Hacker Newszero-daywindowsphysical-accessFeed excerpt
A newly disclosed unpatched zero-day technique called FastUnlock lets someone with physical access to a locked Windows computer start programs without signing in. Whoever uses it gets only a limited account, not administrator control of the machine. The method was posted yesterday by a researcher who uses the handles MSNightmare and Nightmare Eclipse, and the endpoint-security firm ThreatLocker
Researchers show attackers can abuse Gemini Nano, the on-device model in Chrome, as a post-exploitation tool.
arXiv cs.CR (Cryptography & Security)chromegeminipost-exploitationllmFeed excerpt
arXiv:2611.25086v1 Announce Type: new Abstract: Browsers now ship local language models to ordinary endpoints. Google includes Gemini Nano in Chrome to enable productivity features such as summarization, with inference running entirely on the device. These browser-embedded models provide new living-off-the-land opportunities for attackers, each instance a potential post-exploitation engine already installed, signed by the vendor, and invisible to network monitoring. The Gemini model is multimodal, capable of processing files, analyzing screenshots, or conversing with an attacker. We show that it can be driven from the host using only interpreters that ship with the operating system and find that the one-time user-activation check before the model is downloaded is easily satisfied with a simulated click or bypassed entirely.We measure and report the current model's ability to identify credentials on the host, detect PII in email, classify images, and call custom tools. We combine these capabilities and use the model to carry out a simulated cyber operation: reading a password from a webcam image, unlocking an encrypted archive, triaging the contents, and staging a private key for ex…
Cloudflare details five memory corruption vulnerabilities in nginx, called nginx2root, that can allow unauthenticated remote code execution.
Cloudflare Blognginxcvememory-corruptioncloudflareFeed excerpt
Five memory corruption bugs in nginx, called nginx2root, can let an unauthenticated attacker run code on a server. Here’s how they work, and why origins behind Cloudflare's proxy are shielded from the most serious of these bugs.
WSJ reports hackers leaked customer data tied to Visa, Mastercard and JPMorgan, though the summary gives no breach details.
Hacker News (front page)data-leakbreachfinancialFeed excerpt
Article URL: https://www.wsj.com/finance/banking/visa-mastercard-and-jpmorgan-customer-data-leaked-by-hackers-3b5a0f87 Comments URL: https://news.ycombinator.com/item?id=50568082 Points: 30 # Comments: 2
Synology published security advisory SA-26-21 covering vulnerabilities in its products, with the details on its website.
Hacker News (front page)synologyadvisorynasFeed excerpt
Article URL: https://www.synology.com/en-uk/security/advisory/Synology_SA_26_21 Comments URL: https://news.ycombinator.com/item?id=50567209 Points: 75 # Comments: 36
Chase, Visa, and Mastercard confirmed that third-party cyberattacks caused service outages and may have exposed customer information.
The Registerbreachpaymentsthird-partyFeed excerpt
First service outages, then data breaches…what's next?!
Ransomware group claims it stole Nvidia data and set a deadline for payment or leak.
The RegisterransomwarenvidiaextortionFeed excerpt
Ransomware group sets a deadline for the chipmaker to pay up - or see its data leaked
PHP session storage flaw phpAfterShell (CVE-2026-126714) can enable code execution on servers when paired with a gadget chain; fixed in new PHP releases.
The Hacker Newscvephpcode-executionFeed excerpt
PHP's default file-based session storage can be made to return objects an attacker chose, by cutting off a session write at the right moment. When combined with a second weakness, such as a suitable gadget chain in the app, those objects can allow the attacker to run code on the server. Today's PHP releases fix the flaw, called phpAfterShell. The flaw is tracked as CVE-2026-126714 and rated
PHP released security updates fixing 18 vulnerabilities, including critical flaws allowing unauthenticated remote code execution.
BleepingComputerphprcepatchcveFeed excerpt
The PHP development team released security updates addressing 18 vulnerabilities, including critical flaws that could allow unauthenticated attackers to remotely execute code on vulnerable servers. [...]
JPMorgan Chase confirmed a third-party vendor cyberattack exposed customer names and emails, separate from its earlier outage.
The Vergedata-breachransomwarebankingFeed excerpt
JPMorgan Chase says its third-party vendors suffered a cyberattack, exposing marketing data like customer names and email addresses. The confirmation follows a report from Cybernews, which says a ransomware group leaked email addresses belonging to around 16,000 JPMorgan Chase customers. In a post about the attack, JPMorgan Chase says "there is no connection" to the outage that occurred earlier this week, which the company attributed to a "software bug:" We have learned that third parties who provide services to the firm were the target of a cyber attack. Based on our investigation so far, we believe the attackers accessed certain marketi … Read the full story at The Verge.
Norwegian energy company Equinor is recovering from a cyberattack with no timeline, though oil and gas production is unaffected.
The RegistercyberattackenergyincidentFeed excerpt
Norwegian energy giant Equinor says oil and gas production remains unaffected
Critical NGINX 'nginx2root' vulnerabilities allow unauthenticated remote code execution, and can be chained with a Linux kernel bug for root access.
BleepingComputernginxrcelinuxcveFeed excerpt
Two critical NGINX vulnerabilities dubbed "nginx2root" allow unauthenticated attackers to execute code remotely on vulnerable web servers, with researchers demonstrating how the flaws can be chained with a Linux kernel bug to gain root access. [...]
Drupal core patches a critical installer flaw allowing unauthenticated attackers to reinstall sites and possibly execute code.
The Hacker NewsdrupalcverceFeed excerpt
Under specific conditions, a critical vulnerability in Drupal core can allow an attacker without login credentials to use the built-in installer to attempt to reinstall a site. That can lead to a site takeover or remote code execution, and the project rates the vulnerable configurations as uncommon. Fixed releases for all supported branches shipped yesterday. The Drupal Security Team disclosed
PHP 8.5.13, 8.4.28, 8.3.37, and 8.2.36 fix several critical vulnerabilities enabling remote code execution.
LWN.netphprcepatchesFeed excerpt
The PHP Project has announced the release of PHP 8.5.13, 8.4.28, 8.3.37, and 8.2.36. Each contains a number of security and bug fixes, including several fixes for critical vulnerabilities that could allow remote-code execution. According to the project's pre-release announcement on November 18, all of the vulnerabilities of critical severity "require either non-default configuration or uncommon setups". See the release notes for details; Linux vendors and projects should be releasing updated packages soon.
LWN roundup of Friday security advisories from AlmaLinux, Debian, and Fedora covering kernel, Java, MySQL, bind9, and many other packages.
LWN.netsecurity-updatesadvisorylinuxFeed excerpt
Security updates have been issued by AlmaLinux (bind9.18, bubblewrap, fence-agents, freerdp, grafana, grafana-pcp, gstreamer1-plugins-bad-free, jackson-core, jackson-databind, java-17-openjdk, java-21-openjdk, java-25-openjdk, kernel, kernel-rt, mysql8.4, mysql:8.4, qt5-qtbase, qt5-qtdeclarative, raptor2, resteasy, and sos), Debian (electrum, nginx, openjdk-17, pam-mysql, php8.2, and php8.4), Fedora (aws, bind9-next, blender, c-ares, composer, dr_libs, et, gh, incron, libaec, libfontenc, libXfont2, mtools, nebula, perl-Imager, perl-Text-JSCalendar, perl-Text-JSContact, php, python-loguru, python3.6, qpid-proton, roundcubemail, runc, and xen), Gentoo (nginx, PHP, and Ruby), Mageia (hplip), Oracle (.NET 10.0, .NET 11.0, .NET 8.0, .NET 9.0, bubblewrap, emacs, firefox, flatpak-builder, gcc, gimp, glibc, grafana, grafana-pcp, gstreamer1, gvfs, java-1.8.0-openjdk, java-17-openjdk, java-21-openjdk, java-25-openjdk, libssh2, mod_http2, perl-CGI, perl-DBI, perl-DBI:1.641, python3.9, rpm, squid, tesseract, thunderbird, tigervnc, util-linux, webkit2gtk3, and xdg-dbus-proxy), Red Hat (compat-openssl10, kernel, nginx, nginx:1.24, nginx:1.26, nodejs22, nodejs:22, openssl, python-urllib3, python3…
An arXiv paper compares CVSS and EPSS vulnerability prioritization under a fixed CVE analysis budget.
arXiv cs.CR (Cryptography & Security)cvssepssvulnerability-managementFeed excerpt
arXiv:2611.24983v1 Announce Type: new Abstract: Organizations must decide which vulnerabilities warrant further analysis and protective action when the volume of published vulnerabilities exceeds their analytical capacity. We present a methodology for comparing vulnerability-prioritization strategies under an explicit resource constraint, represented by a fixed number of CVEs selected from a common eligible population. We apply this methodology to the Common Vulnerability Scoring System (CVSS) and the Exploit Prediction Scoring System (EPSS). For CVEs published in the National Vulnerability Database (NVD) from 17 March to 31 December 2025 and from 1 January to 31 March 2026 (Q1), we reconstruct each CVE's scores and status 30 days after NVD publication, using archived source data. Each cohort has one ranking per strategy at a 5% budget. Positives are CVEs added to the Known Exploited Vulnerabilities (KEV) catalog within 180 days after their decision day. The catalog is maintained by the Cybersecurity and Infrastructure Security Agency (CISA). EPSS captures 16 of 40 positives in 2025, versus 5 for CVSS; alternative selections among tied CVEs cannot raise CVSS capture above 9. In Q1,…
An arXiv paper shows signed agent-commerce wallet objects can permit transfers beyond the buyer's approved mandate.
arXiv cs.CR (Cryptography & Security)crypto-walletsai-agentsstablecoinsFeed excerpt
arXiv:2611.24989v1 Announce Type: new Abstract: Agent commerce protocols let software agents pay with stablecoins. A buyer approves a payee, asset, spending cap and time window, but the wallet signs an object whose effects depend on the contracts that consume it. Matching the displayed fields or simulating one payment can miss transfers that the signed object still permits. We define authority closure as the set of transfer sequences that holding the signed object can enable. A wallet object respects the mandate when every such sequence meets the buyer's constraints. We characterize when wallet objects can satisfy this property and implement Closure, a wallet-side checker covering 13 families of signed objects. In an evaluation of 2,130 generated objects and 62,979 execution programs, 265 objects with an observed violation passed both a field-and-expiry check and an address allow-list. Under the test model, Closure rejected all 1,680 objects with an observed violation and accepted the remaining 450. In two local checkouts using the x402 reference client, approved payments settled and Closure refused requests outside the mandate before signing. The median decision time was at most 1…
An arXiv paper studies attacks and defenses in decentralized multi-agent LLM systems that delegate tasks between agents.
arXiv cs.CR (Cryptography & Security)llm-agentsprompt-injectionresearchFeed excerpt
arXiv:2611.25056v1 Announce Type: new Abstract: Decentralized multi-agent LLM systems enable independently administered agents, each with its own operator, model, memory, and configuration, to collaborate on complex tasks without centralized coordination. Agents autonomously decompose tasks, execute subtasks and delegate remaining work to peers, enabling scalability and eliminating single points of failure. However, decentralization also introduces significant security challenges, as no single entity has complete visibility into task execution or the authority to enforce correct behavior across agents. In this paper, we systematize threats from malicious agents in decentralized multi-agent systems and propose mitigation techniques. We first design and evaluate attacks against integrity, availability, and confidentiality, demonstrating that a single malicious agent can cause up to 100% task failure, inflate token consumption and runtime by up to 7.7x and 3.2x, respectively, and exfiltrate up to 94% of relevant prior tasks from benign agents. We then present a cryptographic protocol that produces an irrefutable verified execution trace of every action taken in the system and develop …
Proposes forgery-evident undetectable watermarks for generative image models to make provenance schemes resistant to forging attacks.
arXiv cs.CR (Cryptography & Security)watermarkinggenerative-aiforgeryprovenanceFeed excerpt
arXiv:2611.25226v1 Announce Type: new Abstract: Provenance schemes such as those based on content watermarks let images from a generative model be traced back to it. A practical watermarking scheme must be accurate at a low false-positive rate (FPR), robust under content-preserving processing, and undetectable. In addition, it must resist forgery; otherwise, even from a single watermarked image and a surrogate model, an adversary can imprint the watermark onto an arbitrary image or regenerate new content that carries it, falsely attributing harmful content to a model or making authentic photographs appear AI-generated. We argue that a watermarking scheme should not only reject forgeries but also report them: a forgery rejected as unwatermarked leaves no trace of the attempt. State-of-the-art forgery-resilient undetectable watermarking schemes, such as SemBind, fall short on both counts. External anomaly detectors that do report forgeries either miss most of them or falsely flag genuine images after common processing. We propose Tripwire-WM, the first forgery-evident framework for undetectable watermarks. It commits the semantics of the generated image to the magnitudes of the initi…
ChainAgent automatically discovers chainable vulnerabilities in PHP web applications by composing limited flaws into exploit chains.
arXiv cs.CR (Cryptography & Security)phpvulnerability-discoveryweb-securityFeed excerpt
arXiv:2611.25412v1 Announce Type: new Abstract: High-impact attacks on PHP web applications often chain limited-capability vulnerabilities with other flaws or intended behaviors. Discovering these chains requires identifying flaws and the dependencies enabling capability composition. Existing vulnerability discovery techniques analyze flaws in isolation, while attack planning assumes known components. This separation makes it difficult to discover previously unknown components whose security relevance becomes apparent only through their potential roles in an exploit chain. We present ChainAgent, a goal-directed LLM-agent framework that jointly discovers exploit-chain components and their capability dependencies in source code. Guided by an empirical study of 386 real-world web exploit chains, ChainAgent represents vulnerabilities and functional gadgets--intended program behaviors that can be repurposed in attacks--through preconditions and capabilities over security-relevant objects. Instead of discovering vulnerabilities first and constructing attack paths afterward, ChainAgent uses an evolving exploit graph to guide source-code inspection. It searches backward from unmet precondi…
ToolPulse shows cache-based side-channel attacks can eavesdrop on tool invocations of commercial LLM agents running on vendor infrastructure.
arXiv cs.CR (Cryptography & Security)side-channelllm-agentscache-attackFeed excerpt
arXiv:2611.25470v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed as agents that translate user requests into sequences of local tool invocations. While prior cache-based side-channel attacks target the locally deployed models, they typically require the adversary to share model-related components or hardware resources with the victim's inference process. These assumptions do not apply to the predominant agentic settings, in which commercial LLMs execute on vendor-controlled infrastructure. We show that agentic execution exposes a more practical attack surface: although inference remains remote, tools are executed on the user's machine and commonly rely on shared system libraries. In this paper, we first systematically evaluate Flush+Reload and Prime+Probe for monitoring agentic tool invocations. Our evaluation across three hardware platforms shows that Flush+Reload enables accurate, real-time tool monitoring, whereas Prime+Probe, despite supporting cross-VM attacks, incurs prohibitive monitoring overhead and suffers from cache-set aliasing. We present ToolPulse, a cache-based side-channel eavesdropping framework against tool-using agents. We fi…
Tether proposes a verified stateful policy engine that monitors shell and network actions of autonomous AI agents to block malicious effects.
arXiv cs.CR (Cryptography & Security)ai-agentspolicy-enginesandboxingFeed excerpt
arXiv:2611.25599v1 Announce Type: new Abstract: Autonomous AI agents are increasingly deployed for long-running tasks that execute arbitrary shell commands or programs. However, these actions can result in unauthorized network requests or persistent, malicious changes to the user's file system, processes, or privileges. Existing defense mechanisms rely on either LLM-based monitors, which lack provable safety guarantees, or stateless policy enforcement layers, which miss multi-step data flows and execution side-effects unmodeled by the policy language (e.g., network calls initiated inside a Python script). We introduce Tether, a framework for runtime enforcement of stateful security policies over the observed side-effects of an agent's actions. Tether executes each proposed action in an ephemeral, isolated sandbox, inferring abstract action semantics from observed system calls. Tether then enforces stateful policies over these inferred semantics, permitting actions to proceed only if they comply with the policy. Policies and action semantics are defined in an embedded DSL in Lean 4 capable of expressing diverse constraints, including information flow control (IFC), access control, a…
Proposes a migration design letting Ethereum owners enroll post-quantum keys before ECDSA retirement, guarding against Shor-based key extraction.
arXiv cs.CR (Cryptography & Security)post-quantumethereumcryptographyFeed excerpt
arXiv:2611.25663v1 Announce Type: new Abstract: A quantum computer running Shor's algorithm can extract the private key of every Ethereum account whose public key is published, and accounts exposed by their own transactions hold 54.6% of execution-layer ETH. As the first such machine may stay secret, a migration cannot wait for visible thefts, and once keys are extractable, legacy signatures no longer tell owners from attackers. We design a migration in which owners enroll post-quantum keys cheaply with their existing signatures before ECDSA is retired at a scheduled time, while a zero-knowledge proof of the wallet's BIP-39 seed can correct any such enrollment until a common deadline afterward; an account nobody enrolled stays frozen and recoverable by seed proof without a deadline. Each proof binds the new key, a nonce and a fee cap, so neither a paying sponsor nor a reordering block builder can redirect or replay it. Validators switch to hash-based keys first, at a finalized checkpoint and under a stated bound on the attacker's voting power. We prove the account guarantees under explicit assumptions, model-check the rules, and implement the design in go-ethereum and a Lean consen…
Describes BlockJack, a fault attack redirecting paged KV cache block-table translations in vLLM to leak data across tenants.
arXiv cs.CR (Cryptography & Security)llm-servingfault-attackvllmFeed excerpt
arXiv:2611.25945v1 Announce Type: new Abstract: Large language model (LLM) serving frameworks accelerate inference by retaining previously computed states in a key-value (KV) cache. Paged serving frameworks such as vLLM organize this cache through block tables analogous to OS page tables, which map each request's logical KV blocks to physical KV blocks, forming a software-managed address translation layer. Although fault attacks such as Rowhammer have exploited OS page tables to redirect memory accesses, prior fault attacks on LLMs have largely targeted the data plane, i.e., model weights or KV-cache contents, leaving KV-cache mapping unexplored. We present BlockJack (i.e., Block Hijack), the first KV cache address-translation fault attack in LLM inference. By corrupting block-table entries, BlockJack redirects a request's KV cache reads to blocks assigned to another tenant's request without modifying model parameters, prompts, or cached KV states. It achieves both confidentiality violation (victim information appears in attacker responses) and integrity violation (attacker-supplied context steers victim answers). We study paged KV cache mappings in vLLM and formally characterize w…
SoK paper rates LLM agents' software vulnerability discovery and patching capabilities on a driving-automation-style autonomy scale.
arXiv cs.CR (Cryptography & Security)llmvulnerability-researchsurveyFeed excerpt
arXiv:2611.26601v1 Announce Type: new Abstract: LLM-based agents are becoming increasingly capable at discovering, validating and patching vulnerabilities in real software, and vendors and researchers report that this is already changing how vulnerabilities are found and fixed. These claims, however, are scattered across papers, vendor reports and disclosures and are rarely comparable. We introduce a framework that rates these capabilities on an autonomy scale modeled on the levels of driving automation, and qualifies each claim by its correctness, its availability and the strength of its evidence. We use this framework to systematize the published evidence on LLM capability in vulnerability research. We find that agents discovered, validated, and patched vulnerabilities in real-world software without human intervention before the restricted frontier models of 2026, and that open-weight models subsequently attained comparable autonomy for discovery and validation. We then measure the consequences of evolving LLM security capabilities in a decade of public vulnerability records, advisories and exploitation data. In 2026 the number of disclosed vulnerabilities more than doubled, with…
Measuring Hallucination in Large Language Models for Cyber Threat Intelligence: An Exploratory Study3
HalluVision measures hallucinations in LLM-generated cyber threat intelligence summaries and question answers.
arXiv cs.CR (Cryptography & Security)llmthreat-intelhallucinationFeed excerpt
arXiv:2611.25590v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used in cyber threat intelligence (CTI) tasks, such as summarization and question answering. However, they may produce hallucinations, false or misleading content not grounded in the source text, which can mislead analysts and corrupt downstream security decisions. This paper presents an exploratory measurement study of hallucinations in LLM-based CTI. We introduce HalluVision, a reproducible framework that generates LLM outputs, extracts domain-specific entities using a fine-tuned named entity recognition model, and evaluates factual consistency using multiple similarity metrics. Using 4,940 real-world cybersecurity articles, we measure hallucinations in outputs from two commercial LLMs and one open-weight LLM. Measurements are validated against 100 human-annotated outputs and two established faithfulness detectors. Results show that semantic similarity aligns with human judgments, whereas entity-based metrics exhibit limited agreement. Differences between models are concentrated in question answering, while entity extraction quality limits what entity-level metrics can reveal.
Triosecuris is a machine-checked Rocq-verified defense combining CET-style CFI with speculative load hardening against Spectre BTB, RSB, and PHT attacks.
arXiv cs.CR (Cryptography & Security)spectreformal-verificationcontrol-flow-integrityFeed excerpt
arXiv:2601.22978v4 Announce Type: replace Abstract: This paper introduces Triosecuris, a formally verified defense against Spectre BTB, RSB, and PHT that combines CET-style hardware-assisted control-flow integrity with compiler-inserted speculative load hardening (SLH). Triosecuris is based on the novel observation that in the presence of CET-style protection, we can precisely detect BTB misspeculation for indirect calls and RSB misspeculation for returns and set the SLH misspeculation flag. We formalize Triosecuris as a transformation in Rocq and provide a machine-checked proof that it achieves relative security: any transformed program running with speculation leaks no more than what the source program leaks without speculation. This strong security guarantee applies to arbitrary programs, even those not following the cryptographic constant-time programming discipline.
A cross-modal backdoor attack poisons only the connector of connector-based multimodal LLMs, exposing a supply-chain risk in open-source assembled models.
arXiv cs.CR (Cryptography & Security)backdoorllmsupply-chainFeed excerpt
arXiv:2605.07490v2 Announce Type: replace Abstract: Assembling multimodal large language models (MLLMs) from pretrained components has become a widely adopted practice in the open-source ecosystem, yet the supply-chain risks of lightweight connectors remain largely unexplored. We propose a cross-modal backdoor attack that exploits this gap. By poisoning only the connector with a single seed sample and a small set of augmented variants from one modality, the adversary establishes a compact malicious region in the shared latent space associated with a target output. A centroid extracted from this region then serves as a latent anchor for cross-modal activation: adversarial inputs from any other modality can be optimized offline to reach this anchor, triggering the target response without access to the deployed model. Evaluations on PandaGPT and NExT-GPT show same-modality attack success rates (ASR) up to 99.9%, with most cross-modal settings exceeding 95.0% ASR under bounded perturbations. Experiments on ChatBridge further demonstrate the attack's effectiveness across different connector architectures. The attack produces 0.0% backdoor leakage on clean inputs and maintains weight-cos…
The paper proposes behavioral integrity verification to compare declared and actual capabilities of third-party AI agent skills.
arXiv cs.CR (Cryptography & Security)ai-agentssupply-chainverificationFeed excerpt
arXiv:2605.11770v2 Announce Type: replace Abstract: Agent skills extend LLM agents with privileged third-party capabilities such as filesystem access, credentials, network calls, and shell execution. Existing safety work catches malicious prompts and risky runtime actions, but the skill artifact itself goes unverified. We formalize this as the behavioral integrity verification (BIV) problem: a typed set comparison between declared and actual capabilities over a shared taxonomy that bridges code, instructions, and metadata. The BIV framework instantiates this comparison by pairing deterministic code analysis with LLM-assisted capability extraction. The resulting structured evidence supports three downstream analyses: deviation taxonomy, root-cause classification, and malicious-skill detection. On 49,943 skills from the OpenClaw registry, the deviation taxonomy reveals a pervasive description-implementation gap: 80.0% of skills have at least one deviation in the taxonomy, with four compound-threat categories surfaced. Root-cause classification finds that deviations are mostly oversight, not malice: 81.1% trace to developer oversight and 18.9% to adversarial intent, with 5.0% of skill…
Flaws in VMware and KVM need attention, if you have time while fixing Java, PHP, and everything else3
The Register flags critical VMware and KVM vulnerabilities needing attention alongside a wave of Java and PHP patches.
The RegistervmwarekvmpatchesFeed excerpt
Critical Virtzilla bugs, plus a strange Digital Ocean reaction to a nested virtualization issue
The Hacker News weekly roundup covers GPT-6.2 jailbreak flaws, BubbleLoader PNG malware, and $500M ransom demands.
The Hacker NewsroundupmalwarejailbreakransomwareFeed excerpt
The easiest way to fool someone is to give them something they already expect. A job offer. A message about a service outage. A familiar app. Nothing looks strange at first. And that's the problem. This week's stories show how ordinary things are being used to steal data, spread malware, and get people to hand over access. There's another shift worth watching. AI is making some attacks easier
Transport for London disclosed a data breach that may have exposed account details of about 265,000 Oyster photocard users.
BleepingComputerbreachtfldata-leakFeed excerpt
Transport for London (TfL) has announced that around 265,000 Oyster photocard customers may have had their account details exposed in a data breach. [...]
US prosecutors allege a Russian national stalked a San Francisco techie in a four-month operation to steal defense-tech secrets.
The RegisterespionagearrestdefenseFeed excerpt
DoJ claims suspect part of 4-month operation to steal secrets, surveil employees
Roundup of security updates across AlmaLinux, Debian, Fedora, Mageia, Slackware, and SUSE distributions.
LWN.netsecurity-advisorylinuxpatchesFeed excerpt
Security updates have been issued by AlmaLinux (cockpit, cockpit-files, cockpit-machines, gimp, glib-networking, krb5, libpcap, nginx, nodejs26, perl-CGI, php8.4, skopeo, tigervnc, and xorg-x11-server), Debian (haproxy, libreoffice, libvisio, needrestart, openjdk-21, and openjdk-25), Fedora (asterisk, erlang, ffmpeg, ignition, libosmocore, luksmeta, msedit, openapv, passwordsafe, rabbitmq-server, sngrep, and vyper), Mageia (firefox, nss, rootcerts, rsync, and thunderbird), Slackware (libfontenc and libXfont2), SUSE (389-ds, apache2, aspell, bind, brotli, busybox, civetweb, containerd, cosign, docker, gleam, glibc, graphicsmagick, helm, ignition, ImageMagick, java-17-openjdk, java-21-openjdk, java-25-openjdk, jhead, kernel, kubernetes, libsofia-sip-ua-glib3, mcphost, perl-HTTP-Body, perl-MIME-tools, phodav, python-asyncssh, python-django, python-joserfc, python315, qemu, rootlesskit, roundcubemail, rsyslog, runc, tempo-cli, trivy, and xorg-x11-server), and Ubuntu (cloud-init, gstreamer1.0, hplip, libcpanel-json-xs-perl, linux, linux-aws, linux-aws-5.4, linux-aws-fips, linux-azure-5.4, linux-azure-fips, linux-bluefield, linux-fips, linux-gcp, linux-gcp-5.4, linux-gcp-fips, linux-hwe-…