Intelligence Feed

1mo ago · hacker-news

Cisco Talos identified a novel Windows malware named CLOSEDQUORUM that leverages commercial AI models—DeepSeek, Qwen, Mistral, and Google Gemini—to make autonomous decisions on attack actions through a voting mechanism. The malware performs credential theft, code injection, persistence, and lateral movement based on consensus from AI models, with decisions logged and data exfiltrated via a Discord webhook. While the public version is non-functional due to placeholder API keys and webhook URLs, the malware represents a new paradigm in AI-driven threats. Indicators include specific behaviors such as AI service queries from non-AI applications, LSASS memory dumping, process injection, and WMI-based persistence with Windows Update-themed names.

1mo ago · hacker-news

Threat actors, attributed to North Korea, are distributing Go-based malware through malicious Terraform providers and Go modules hosted on the HashiCorp and npm registries, marking a novel supply chain attack vector. The malware uses dual command-and-control channels via Ethereum smart contracts on testnets and Slack, with execution often gated by cryptographic checks to hinder analysis. The same infrastructure and techniques are linked to the previously known Graphalgo and PolinRider campaigns, indicating a coordinated, targeted operation using fake job interviews as an initial access vector.

1mo ago · bleeping-computer

A financially motivated threat actor is leveraging AI-powered tools to conduct large-scale attacks on online retailers, stealing over 600,000 credit card records and deploying skimmer malware on at least 119 websites. The campaign uses three AI tools—Strix for vulnerability scanning, Cairn for autonomous exploitation, and Hermes for campaign orchestration with decision-making powered by claude-opus-4.6. The attacker, believed to be Chinese, provides high-level instructions while the AI agents execute attacks, including injecting skimmers via multiple methods and performing post-exfiltration data cleanup to erase traces in databases, causing operational disruptions.

1mo ago · bleeping-computer

Threat actors are actively exploiting a critical unauthenticated path traversal vulnerability in WordPress, tracked as CVE-2026-87902, to achieve remote code execution under specific conditions. The exploitation involves writing malicious PHP files to the server's filesystem, which execute shell commands when accessed. Initial activity began within hours of the patch release, with reconnaissance escalating to active payload delivery, including file writing to /tmp and /var/tmp directories using attacker-controlled content.

1mo ago · bleeping-computer

Check Point has confirmed active exploitation of two critical vulnerabilities in its Security Gateway product: CVE-2026-85102, a pre-authentication remote code execution flaw in the VPN certificate-handling functionality, and CVE-2026-93616, a pre-authentication path traversal vulnerability in the Management web service. Exploitation of CVE-2026-93616 began as a zero-day on July 23, 2026, and widespread attacks on CVE-2026-85102 started on September 12, with threat actors using anonymizing infrastructure such as VPNs and proxies. The U.S. CISA has added both flaws to its Known Exploited Vulnerabilities catalog, requiring federal agencies to patch by September 25, 2026.

1mo ago · bleeping-computer

A new Android malware-as-a-service (MaaS) platform named RemControl is targeting users in Europe, Canada, and the Middle East through malvertising campaigns impersonating the TVTap IPTV app. The malware uses phishing overlays to steal banking credentials, leverages Accessibility Services for persistent access, and can stream real-time UI data to attackers. It communicates with C2 servers via Telegram channels and uses a VPN service to bypass Google Play Protect checks.

1mo ago · bleeping-computer

The domain third-party.com, commonly used as a placeholder in developer documentation, is now being used to serve a ClickFix attack that mimics a Cloudflare security verification page. When users click a 'Verify you are human' prompt, a malicious PowerShell command is copied to the clipboard, instructing them to paste and execute it via the Windows Run dialog. The command downloads and runs a PowerShell script that fetches a 134MB zip archive named update26.zip, which extracts and attempts to execute a file called draw.io.exe. The attack specifically targets Windows users, while macOS and Linux users are shown an error message and not served the malicious payload.

1mo ago · bleeping-computer

A security vulnerability dubbed 'ConfigConfusion' allows attackers with access to a Kubernetes namespace to escalate privileges and gain full control of a Google Cloud (GCP) organization by exploiting misconfigurations in Google Kubernetes Config Connector (KCC). KCC uses a single, highly privileged service account to manage cloud resources on behalf of developers, but does not validate whether the requesting user has appropriate Google Cloud IAM permissions. As a result, an attacker can submit a malicious IAMPolicyMember YAML resource to grant themselves owner-level access to the entire GCP organization, even without possessing any cloud credentials. This represents a confused deputy problem where KCC acts as an overprivileged intermediary.

1mo ago · hacker-news

A use-after-free vulnerability in the Linux kernel's AF_UNIX socket subsystem, tracked as CVE-2026-80521, enables container escape to gain root privileges on the host. The flaw affects unpatched Ubuntu 26.04, 24.04, and 22.04 LTS releases, despite an upstream fix being available since August 6, 2026. Security firm DepthFirst released exploit code targeting Ubuntu 26.04, demonstrating the practical risk of containerized environments where default seccomp policies allow access to AF_UNIX sockets. Although no active attacks have been confirmed, the vulnerability undermines container security assumptions, especially as AI-assisted research accelerates discovery of such flaws.

1mo ago · hacker-news

cPanel disclosed three security flaws, two of which are critical, allowing privilege escalation and unauthorized access on shared hosting servers. CVE-2026-87899 enables a logged-in cPanel account holder to execute code as root via the CalDAV and CardDAV service, achieving full server control. CVE-2026-87900 affects the WP Toolkit plugin, allowing a user to modify databases belonging to other accounts. CVE-2026-68490 permits local users to read calendar and contact data from other accounts, though not modify it or escalate privileges. All vulnerabilities have been patched, but no exploitation in the wild has been reported.

1mo ago · hacker-news

Unknown threat actors have compromised specific versions of the MemTensor packages on npm and PyPI, injecting a malicious Go-based credential stealer named sckit. The malicious packages, when used in development or CI environments, execute a payload that harvests sensitive credentials from cloud services, source control, package registries, and developer tools. The malware exfiltrates stolen data to the domain skyleen[.]fr and can self-propagate via GitHub, npm, and PyPI. Organizations are advised to pin to known clean versions, rotate secrets, and block the C2 domain.

1mo ago · bleeping-computer

Arista Networks has patched a zero-day vulnerability, tracked as CVE-2026-93952, in its VeloCloud Orchestrator (VCO) On-Prem deployments that was actively exploited in the wild. The flaw stems from improper input validation and allows remote unauthenticated attackers to access privileged internal functionality by exploiting certificate-based authentication mechanisms. The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has added the vulnerability to its Known Exploited Vulnerabilities catalog, requiring federal agencies to remediate by September 25, 2026.

1mo ago · hacker-news

ShinyHunters, a cyber extortion group, claimed responsibility for breaching the U.S. Federal Bureau of Investigation (FBI), asserting they exfiltrated sensitive data on current and former agents and job applicants. The group reportedly exploited a zero-day vulnerability in Oracle PeopleSoft to gain remote code execution and deface the FBI's jobs portal, FBIjobs.gov, with a taunting banner. They linked the attack to retaliation for an FBI public service announcement criticizing their prior activities, particularly around the Canvas LMS breach. While the FBI has acknowledged awareness of the claims and is investigating, no technical evidence or data dumps have been independently verified yet.

1mo ago · hacker-news

A critical vulnerability in Next.js, tracked as CVE-2026-94545, allows server-side code execution when attacker-controlled input is processed by the ImageResponse feature during SVG generation. The flaw exists in versions 16.2.0 through 16.3.5 when using the Node.js runtime, where unsanitized input can be interpreted as SVG code due to improper escaping in the underlying Satori library. This could enable remote attackers to execute arbitrary code on the server if user-supplied values are embedded in SVG content, attributes, or styles. The vulnerability was patched in Next.js 16.3.6 and Satori 0.33.5, with no public exploits or active attacks reported at the time of disclosure.

1mo ago · hacker-news

A Chinese threat actor, UTA0565, exploited a zero-day chain involving vulnerabilities in Google Chrome and Microsoft Windows to deploy a new malware family named CLEANGULP. The attacks, observed on September 3 and 4, 2026, used fake websites mimicking media organizations and NGOs to deliver the BlueMoon exploit kit, which leveraged CVE-2026-85046, CVE-2026-87491, and CVE-2026-85880 to escape the browser sandbox and execute code remotely. The malware, delivered as 'chrome_cleanup.exe', communicates with a hard-coded C2 domain and provides capabilities including command execution, process listing, file upload/download, and beacon object file execution.

1mo ago · hacker-news

F5 has patched a critical zero-day vulnerability, CVE-2026-94127, in its BIG-IP Access Policy Manager (APM) that allows unauthenticated remote code execution on systems where APM is configured as an OAuth authorization server. The vulnerability is a heap-based buffer overflow with a CVSS v3.1 score of 9.8, and it is actively exploited in the wild. Malicious traffic targeting the virtual server can trigger remote code execution, bypassing protections on the management interface. The U.S. CISA has added the flaw to its Known Exploited Vulnerabilities catalog and mandated federal agencies to apply mitigations within three days.

1mo ago · bleeping-computer

F5 has released security updates to address a critical zero-day vulnerability, CVE-2026-94127, in its BIG-IP APM (Access Policy Manager) product that is being actively exploited in remote code execution attacks. The vulnerability affects systems configured as an OAuth Authorization Server with specific access policies and OAuth profiles on a virtual server. Exploitation can lead to remote code execution, and F5 has confirmed active attacks. The Cybersecurity and Infrastructure Security Agency (CISA) has added the vulnerability to its Known Exploited Vulnerabilities catalog and mandated federal agencies to remediate by a specified deadline.

1mo ago · talos

CLOSEDQUORUM is a Windows-based malware implant that uses commercial large language models (LLMs) from DeepSeek, Qwen, Mistral, and Google Gemini as an autonomous command and control (C2) infrastructure. It delegates tactical decisions such as credential theft, process injection, and persistence to a panel of LLMs, which vote on the next action via a structured JSON schema. The malware targets user credentials and cryptocurrency wallets, exfiltrating stolen data via an operator-controlled Discord webhook using AES-256-GCM encryption and Base64 encoding. While the distributed version contains placeholder keys and is non-functional, development builds suggest a 'credentials-as-a-service' model where customized binaries are provided to operators.

1mo ago · hacker-news

A former Microsoft security researcher, Abdelhamid Naceri, publicly released a zero-day proof-of-concept tool called BigDiskBuster that blocks Microsoft Defender from installing platform and signature updates by consuming all available disk space. The tool monitors Defender's update paths and creates a large hidden temporary file to prevent update downloads, deleting it once the update attempt fails. This denial-of-service technique could leave Defender with outdated detection capabilities, increasing risk of undetected malware. No patch or CVE has been issued, and the tool reportedly works across all supported Windows versions.

1mo ago · hacker-news

A critical vulnerability in Bifrost, an open-source AI gateway, allows unauthenticated attackers to execute arbitrary commands on the gateway server via a single HTTP request. The flaw, tracked as CVE-2026-90898, affects all versions prior to 2.1.0 when management authentication is disabled by default. Attackers can exploit the /api/mcp/client endpoint to register a stdio-type MCP client, leading to immediate command execution as the gateway process user, potentially exposing stored API keys for LLM providers. A second related vulnerability, CVE-2026-86242, allows unauthenticated registration of a custom plugin from an HTTP URL, resulting in remote code execution or server-side request forgery depending on the build type.

1mo ago · hacker-news

Microsoft, in collaboration with multiple public and private partners, disrupted the EvilTokens phishing-as-a-service (PhaaS) platform, which leveraged AI to automate and scale business email compromise (BEC) attacks. The service abused OAuth 2.0 device authorization flows to gain persistent access to Microsoft accounts without stealing passwords, using social engineering to trick users into entering device codes on legitimate Microsoft login pages. EvilTokens offered AI-powered tools to analyze compromised inboxes, identify financial workflows, and draft convincing phishing emails, enabling even low-skill attackers to conduct sophisticated fraud. The operation led to the seizure of 50 websites, takedown of 150 domains, and the arrest of two individuals, with evidence linking the service to over 12,000 compromised inboxes across 10,000 organizations worldwide.

1mo ago · hacker-news

A malicious npm package named 'tw-pkgprobe-7731' was uploaded to the npm registry by a now-defunct account 'twdepprobe7731', masquerading as an authorized Twilio bug-bounty probe. The package targets Twilio developers by checking for specific environment conditions and, if met, exfiltrates environment variables, system configurations, and Twilio credentials such as ACCOUNT_SID and AUTH_TOKEN. Later versions attempted to inject malicious npm packages and perform OSINT gathering on Twilio-related infrastructure, though the final versions reverted to benign behavior, suggesting possible testing or evasion. The package violates Twilio's HackerOne security research guidelines and shows no obfuscation, indicating a low sophistication threat actor.

1mo ago · hacker-news

WordPress has patched a critical vulnerability, CVE-2026-87902, that allows unauthenticated attackers to exploit a path traversal flaw in the template loading mechanism, potentially leading to remote code execution on vulnerable servers. The vulnerability affects all WordPress versions from 4.7.0 to 7.1.1 and stems from insufficient sanitization of user-supplied input used to construct template file paths. Successful exploitation requires the active theme to have a top-level directory starting with 'page-' and depends on server-side PHP configuration (e.g., register_argc_argv enabled) for full code execution. While no active exploitation has been observed as of September 22, 2026, the flaw is rated with a CVSS score of 9.2 and affects a wide range of WordPress installations.

1mo ago · bleeping-computer

Check Point has released emergency hotfixes for a critical zero-day vulnerability, CVE-2026-93616, in its Security Management Server that allows unauthenticated attackers to perform path traversal and upload and execute arbitrary scripts. The vulnerability is actively exploited in the wild, with Check Point confirming attacks against a handful of customers. The affected products include Security Management Server, Multi-Domain Security Management Server, Log Server, Multi-Domain Log Server, and SmartEvent. CISA and FBI have previously warned software vendors about the persistence of such path traversal flaws, which remain a high-risk attack vector.

1mo ago · bleeping-computer

Cisco Talos discovered a new Windows malware named ClosedQuorum, which autonomously makes post-compromise attack decisions using multiple AI models including Google Gemini, DeepSeek, Qwen, and Mistral. The malware uses a voting system among the AI models to decide actions such as credential theft, code injection, persistence, and lateral movement, with DeepSeek breaking ties. Stolen data is exfiltrated via Discord webhooks. While the sample analyzed contains placeholder credentials and no active lateral movement capability, it represents an early example of fully automated, AI-driven attack chains with no human operator involvement.

1mo ago · bleeping-computer

The ShinyHunters extortion gang claims to have breached FBI systems using an unpatched zero-day vulnerability in Oracle PeopleSoft, enabling remote code execution and lateral movement into FBI-managed AWS GovCloud infrastructure. They allege exfiltration of 2–3TB of sensitive data, including personally identifiable and health-related information of current and former FBI employees and job applicants. ShinyHunters defaced the FBI Jobs website (apply.fbijobs.gov) with their branding and claim to be exploiting the same vulnerability against Fortune 500 companies. The group stated the attack was retaliation for an FBI FLASH report published in May 2026, and they have not ruled out releasing the stolen data.

1mo ago · hacker-news

A vulnerability in SharePoint Server, tracked as CVE-2026-65660, initially misclassified by Microsoft as a spoofing flaw, actually enables authenticated remote code execution. The flaw affects SharePoint Server 2016, 2019, and Subscription Edition, and stems from improper handling of double quotes in web-part markup within the ToolPane component, allowing arbitrary .NET class registration and subsequent code execution via XamlServices.Parse() deserialization. Researcher Dinh Ho Anh Khoa demonstrated the exploit and noted it could be chained with a previously patched authentication bypass to achieve pre-authentication RCE on anonymously accessible servers. Although no in-the-wild exploitation has been reported, the full exploit is public, and the patch was released in Microsoft's August 11 updates.

1mo ago · hacker-news

A vulnerability in the Linux kernel's KVM virtualization code for ARM64 processors, tracked as CVE-2026-89775, allows a guest virtual machine to read and write host kernel memory, potentially enabling guest-to-host escape. The flaw affects systems with nested virtualization enabled and stems from a missed TLB invalidation due to a zero-size calculation when handling guest memory. It impacts Linux kernels starting from version 6.17 and has been patched in 6.18.51, 7.2.5, and 7.3-rc1. While no active exploitation has been observed, local users on vulnerable systems—such as those with /dev/kvm accessible—could exploit it to escalate privileges to root.

1mo ago · hacker-news

A critical CVSS 10.0 vulnerability, CVE-2026-93952, in on-premises VeloCloud Orchestrator (VCO) is actively being exploited by attackers. The flaw allows unauthenticated remote attackers to escalate privileges and compromise the VCO host, but only affects deployments using certificate-based authentication from Edge devices to the orchestrator. Arista has released patches for the 5.2 and 6.4 release trains, but fixes for 6.1 and 7.0 are pending. Indicators of compromise include specific files, an MD5 hash, suspicious IPs, and an HTTP header in nginx logs. Organizations are advised to limit access to the VCO web interface, monitor for malicious activity, and apply patches as soon as available.

1mo ago · bleeping-computer

D-Link has issued a warning about a maximum-severity zero-day vulnerability, CVE-2026-86296, affecting legacy DIR-822A Wi-Fi routers. The flaw is a stack-based buffer overflow in the DHCP server component (udhcpcd/serverpacket.c) that can be exploited without authentication or user interaction by sending crafted DHCP packets. Proof-of-concept exploit code has been publicly released, increasing the risk of weaponization. D-Link is also investigating a second critical vulnerability, CVE-2026-86510, an out-of-bounds write in the L2TP control message parser, which could allow arbitrary memory corruption on devices using L2TP or L2TPv6 WAN connectivity.