ai infrastructure

2026-07-169 minMandiantlow

This article provides a structural blueprint for safely integrating LLM agents into vulnerability management workflows, covering both enterprise vulnerability management and product security tracks. It outlines operational guardrails including pre-agent data security, workload isolation, least-privileged machine identities, toxic flow analysis, and supply chain resilience for AI skills. The guidance emphasizes that LLMs augment but do not replace deterministic controls, human threat modeling, and secure-by-design principles, and recommends phasing memory-safe languages into new development as a long-term strategy.

2026-07-1610 minZscaler ThreatLabzhigh

A MacSync Stealer campaign abuses Anthropic's Claude shared chat platform to host ClickFix-style malicious instructions, using Google malvertising to lure macOS users searching for Claude. Victims are tricked into running a Base64-obfuscated curl command that pipes a zsh script to execution, which in turn downloads and pipes a second stage to osascript. The stealer collects keychain data, browser credentials, cryptocurrency wallets, cloud keys, and user files, exfiltrating them in 10MB chunks via HTTP PUT before deleting all traces.

2026-07-1410 minAkamaicritical

Attackers are actively exploiting CVE-2025-3248, a critical RCE vulnerability in Langflow's code validation API, to deploy a customized Gafgyt DDoS bot on AI development servers. The bot uses a modified RC4 stream cipher for C2 communications and is optimized purely for network flooding attacks (UDP, TCP, HOLD, Junk). AI infrastructure is targeted due to high bandwidth availability, shadow IT deployment practices, and permissive egress filtering.

2026-07-145 minCheck Pointhigh

The Check Point Research AI Security Report 2026 highlights the transition of AI from an attack assistant to a live attack operator. Threat actors are now using AI to build deployment-ready malware, run live intrusions, and scale social engineering attacks using forged virtual identities. The report also notes a significant rise in indirect prompt injection attacks and persistent enterprise data leakage through GenAI applications.

2026-07-139 minCheck Pointhigh

This weekly threat intelligence bulletin covers multiple significant incidents including autonomous LLM-driven ransomware (JadePuffer), a cryptocurrency supply chain compromise via malicious npm packages, and three critical CVEs affecting Langflow, Tenda routers, and Linux KVM. Iran-linked Cavern Manticore and China-linked UAT-7810 were profiled targeting Israeli and networking infrastructure respectively. The report also highlights risks in AI development tools where hidden malicious instructions in open-source files could achieve RCE through Claude Code and OpenAI Codex.

2026-07-0915 minSockethigh

Operation Muck and Load is a supply chain attack campaign centered on a malicious Go module that impersonates a DNS scanner tool to deliver a multi-stage Windows malware loader. The campaign leverages 222 GitHub lure repositories across 190 accounts with automated commit-farming workflows to create false credibility, and uses public dead-drop resolvers across multiple platforms for resilient payload-location resolution. The final payload chain delivers AsyncRAT, Quasar, Remcos, Vidar infostealer, and cryptominers through password-protected 7z archives extracted into masqueraded Microsoft-themed directories.

2026-07-079 minVaroniscritical

Varonis Threat Labs discovered a critical vulnerability in Google Cloud Platform's Dialogflow CX service that allowed attackers with only the dialogflow.playbooks.update permission to inject persistent malicious code into the shared Cloud Run execution environment. The vulnerability enabled silent exfiltration of conversation data, bypassing VPC Service Controls, and injection of phishing prompts into chatbot conversations. Additional weaknesses included IMDS credential exposure and unrestricted outbound network access from the Cloud Run environment.

Token Theft and AI Poisoning Redefine the Perimeter Attackers are shifting from breaking passwords to stealing active login sessions, bypassing multi-factor authentication entirely. This week, ARToken and ConsentFix exploited Microsoft 365 OAuth flows to hijack accounts, while Anubis ransomware used the ongoing CitrixBleed 2 vulnerability to steal session tokens from network gateways. Even a standard user can become a Global Administrator in minutes if identity settings are loose, as demonstrated by a recent M365 privilege escalation analysis. Simultaneously, artificial intelligence systems have evolved from helper tools to critical vulnerabilities, serving as both the weapon and the target. Threat actors are using AI to generate malware like InfernoGrabber v9.0 and BusySnake Stealer, while also poisoning AI agent ecosystems with malicious skills like OpenClaw and tricking AI models into executing financial fraud via indirect prompt injection. The AI arms race has accelerated breakout times to under 30 minutes, with state-sponsored groups like GTG-1002 now orchestrating entire espionage campaigns via AI. Defenders must immediately audit identity and session controls, treating session tokens as highly sensitive credentials. Security teams should also implement guardrails for AI agents, verifying external URLs and restricting autonomous financial or code execution actions.

AI is not fundamentally changing adversary capabilities but is compressing attack timelines, lowering operational costs, and scaling existing tactics. Breakout times have dropped to an average of 29 minutes, with AI-enabled operations increasing 89% year-on-year. The most significant emerging threats are runtime-LLM malware (PROMPTSTEAL/LAMEHUG, QUIETVAULT) that query language models during execution, and agentic AI operations (GTG-1002) where AI agents conduct multi-stage intrusions with minimal human steering. Defenders face a dual pressure: faster attacks and an expanding attack surface from AI supply-chain dependencies.

OpenClaw, a widely adopted AI agent ecosystem with a skill marketplace called ClawHub, is being actively abused by attackers who publish malicious skills containing embedded shell commands or harmful natural-language instructions. Approximately 530 vulnerabilities have been discovered in OpenClaw and its underlying technologies, many involving sensitive data storage and excessive privileges. Kaspersky identified over 600 malicious skills from 24 accounts in April, with 1,100+ malicious accounts created since January. Malicious skills observed include macOS payloads using base64-encoded curl-pipe-to-bash execution and Windows MSI installers distributed via password-protected ZIP archives.

AI Attacked and Abused While Perimeter Authentication Collapses The month's defining shift was the emergence of AI as a two-sided battlefield: organizations deployed AI tools faster than they secured them, while attackers weaponized the same technology against defenders. Critical flaws in LangGraph allowed SQL injection chained to remote code execution, M365 Copilot could be turned into a one-click data exfiltration weapon via SearchLeak, and Langflow was exploited to deploy cryptominers. Meanwhile, the ongoing Shai-Hulud campaign injected prompts to blind AI malware scanners, macOS.Gaslight turned prompt injection against human analysts, and Russia's APT28 began experimenting with LLM-integrated malware. At the same time, perimeter authentication collapsed at scale: FortiBleed exposed credentials for over 73,000 FortiGate firewalls, CVE-2026-50751 let attackers bypass Check Point VPN authentication entirely, and ShinyHunters exploited an Oracle PeopleSoft zero-day across over 100 organizations. Supply chain attackers followed developers to their new AI tools, compromising the ecosystems where code is written and built. The Shai-Hulud/Miasma worm expanded from npm into PyPI and injected persistent backdoors into AI coding assistant configurations, while North Korea's Sapphire Sleet compromised over 140 Mastra npm packages to steal cryptocurrency wallets, and the ongoing GlassWorm campaign pivoted to WebAssembly malware in VS Code extensions using the Solana blockchain as command-and-control. Social engineering also industrialized: the ErrTraffic framework turned ClickFix deception into a Malware-as-a-Service operation with blockchain dead drops, and EvilTokens hid phishing flows inside browser-side encryption to defeat network scanners while hijacking Microsoft device-code authentication. Organizations should treat AI deployments as untrusted perimeter assets—restrict their network access, audit third-party skills and extensions, and assume prompt-injection attacks will target both automated scanners and human analysts. Every internet-facing VPN, firewall, and edge appliance should be patched immediately, with credentials rotated and phishing-resistant MFA enforced, because perimeter authentication failures now cascade directly into internal network compromise.

2026-06-246 minPalo Alto Networkshigh

Threat actors are exploiting the OpenClaw AI agent ecosystem by publishing malicious skills on the ClawHub marketplace. These skills leverage semantic instruction hijacking to bypass traditional security controls, delivering macOS infostealers via base64-encoded droppers, utilizing massive file padding for defense evasion, and executing novel agentic financial fraud schemes like runtime affiliate injection and front-running.

2026-06-247 minTrend Microhigh

A cryptocurrency-mining campaign is actively exploiting CVE-2026-33017, an unauthenticated RCE vulnerability in Langflow, to deploy the lambsys malware. The attack chain involves a bash dropper that establishes SSH-based lateral movement, followed by a Go-based payload that systematically disables Linux security controls, eliminates rival miners, and deploys a customized XMRig miner.

Trust Chains Broken at Scale While ClickFix Becomes a Service This week, attackers stopped trying to kick down the front door and instead walked in through the trust chains that hold digital ecosystems together. North Korea's Sapphire Sleet compromised over 140 Mastra npm packages through a single typosquatted dependency, stealing cryptocurrency wallets and planting persistent backdoors on developer machines. The GlassWorm group trojanized Open VSX extensions with WebAssembly malware that uses the Solana blockchain as an unkillable command channel, while SmartApeSG hijacked the Okendo Reviews widget to serve malicious prompts on thousands of e-commerce sites. Even vendor integrations became a liability: the Klue breach exposed Recorded Future client data through a compromised OAuth token connecting a marketing tool to Salesforce. Deception also became an industrial product. The ErrTraffic framework now operates as full Malware-as-a-Service, using blockchain smart contracts to hide its infrastructure and compromised WordPress sites to serve fake error prompts that trick users into running malicious commands. Attackers weaponized trusted AI platforms too—one campaign abused claude.ai's shared chat feature to deliver MacSync infostealer on macOS, while the shai_hulululud npm package uses prompt injection to blind AI-powered security scanners. On the infrastructure side, the FortiBleed campaign cracked credentials for over 73,000 FortiGate firewalls with a 45-GPU cluster, handing attackers valid keys to government and defense networks worldwide. Defenders should immediately hunt for the easy-day-js dependency in their npm projects, reset credentials on any FortiGate firewall, enable Azure AD Graph Activity Logs to close a years-long reconnaissance visibility gap in Microsoft cloud environments, and audit OAuth tokens on all third-party vendor integrations.

2026-06-183 minCisco Talosinfo

The article details a novel approach to reverse engineering where the VB6 disassembler vbdec exposes its internal object model via the Windows Component Object Model (COM). This allows local AI agents to programmatically query and automate complex analysis tasks, such as decompilation and call graph generation, without requiring built-in AI features or cloud uploads.

2026-06-187 minTrend Microhigh

A sophisticated malvertising campaign abused Google Ads to deliver ClickFix social engineering attacks disguised as popular AI developer tools and Mac utilities. The threat actors initially leveraged GitLab Pages before pivoting to weaponize claude.ai's shared chat feature, tricking macOS users into executing terminal commands that deploy the MacSync infostealer.

2026-06-173 minNCSChigh

The NCSC CEO reported that approximately 75% of the 200+ cyber incidents affecting UK critical national infrastructure over the past year were linked to hostile state actors such as Russia, China, and Iran. The NCSC warns that unpatched legacy systems pose a severe risk, particularly as AI-enabled cyber capabilities are projected to accelerate the exploitation of known vulnerabilities at scale by 2028.

2026-06-167 minSekoia.iohigh

ErrTraffic is a Malware-as-a-Service framework that compromises WordPress sites and uses malvertising to deliver ClickFix social engineering lures. It leverages EtherHiding via Polygon smart contracts to dynamically resolve C2 infrastructure and distribute infostealers, RATs, and loaders.

2026-06-165 minPalo Alto Networkscritical

A critical vulnerability in the Google Cloud Vertex AI SDK for Python allows attackers to achieve cross-tenant Remote Code Execution (RCE) via bucket squatting. By predicting default staging bucket names and exploiting a lack of ownership verification, attackers can intercept model uploads and inject malicious pickle payloads, leading to the theft of highly privileged service account tokens.

Perimeter Auth Collapse and AI-Driven Deception Shift the Battlefield The security perimeter cracked open this week as critical authentication bypasses in Check Point VPNs, Ivanti Sentry, and Palo Alto GlobalProtect gave attackers a free pass into corporate networks, with Qilin ransomware already exploiting one to launch real attacks. At the same time, AI became the year's most versatile weapon: criminals used ChatGPT and Claude brands as phishing lures, researchers proved AI email assistants will hand over corporate secrets to impersonators, and the Shai-Hulud campaign began injecting fake prompts to blind AI-powered security scanners. Patch edge VPN appliances immediately, treat AI agents as high-risk insiders, and hunt for device-code authentication events that bypass normal credential checks.

2026-06-136 minZscaler ThreatLabzcritical

The Shai-Hulud software supply chain campaign has significantly evolved, expanding from npm to PyPI and shifting from maintainer compromise to CI/CD abuse. Recent waves demonstrate advanced techniques including OIDC token scraping to bypass SLSA provenance, IDE configuration file weaponization, and prompt injection designed to evade LLM-based security scanners.

2026-06-115 minPalo Alto Networkshigh

The article introduces Behavioral Integrity Verification (BIV) to audit third-party skills for AI agents by comparing declared metadata against actual executable code and natural-language instructions. Analysis of the OpenClaw registry found that while most deviations are benign documentation errors, a critical 5% of skills contain multi-stage attack chains such as silent credential exfiltration and instruction-override hijacking.

2026-06-085 minMorphiseccritical

The third wave of the Shai-Hulud supply chain worm, dubbed Miasma, targets the npm ecosystem by utilizing weaponized binding.gyp files to bypass lifecycle script monitoring. It establishes deep persistence within AI assistant and IDE configuration directories, evades detection through dormancy and EDR checks, and abuses valid Sigstore attestations to masquerade as legitimate packages.

Trojanized Build Pipelines and Blind-Spot Appliances Redefine the Perimeter Attackers are bypassing traditional network defenses by compromising the tools developers use to build software and the AI assistants they rely on to write code. Campaigns like Mini Shai-Hulud and Miasma - The Spreading Blight flooded package registries with malicious code that steals cloud credentials and CI/CD tokens, while researchers proved that public AI agent skill marketplaces are completely ineffective at catching malicious add-ons. Nation-state actors and cybercriminals are simultaneously shifting their focus to blind spots in corporate networks and trusted platforms. The VerdantBamboo group exploited firewalls to bypass conditional access, while UNC3753 used IT impersonation to trick law firm employees into installing remote access tools, and Kali365 expanded its phishing infrastructure to steal multi-factor authentication tokens. Defenders must shift their focus from perimeter email filtering to securing the software build pipeline and monitoring edge appliances for anomalous traffic. Hunt for unexpected connections to cloud storage APIs and review developer environments for compromised packages or AI skills.

2026-06-077 minSocketcritical

A coordinated supply chain attack compromised 19 PyPI packages, utilizing malicious .pth files to achieve execution at Python startup. The loader downloads the Bun runtime to execute an obfuscated JavaScript stealer targeting developer secrets, cloud credentials, and CI/CD tokens, exfiltrating data via GitHub repositories and Actions.

2026-06-044 minANY.RUNmedium

ANY.RUN's Q1 2026 Cyber Risk Report highlights a significant acceleration in attacker operational tempo, with the median time-to-persistence dropping to 21 seconds and LOTL execution occurring in 16 seconds. The data also shows a marked increase in loader-based attacks, credential theft, and the weaponization of trusted tools via JavaScript LOLBAS techniques, emphasizing the critical need for rapid, behavior-based detection capabilities.

Session Hijacking and Developer Tool Poisoning Collapse Authentication Trust This week, attackers proved that multi-factor authentication is no longer a reliable gatekeeper. Campaigns like Tycoon 2FA and Chinese-language PhaaS platforms intercept one-time passwords in real time and steal session tokens to maintain persistent access, while infostealers like EKZ Infostealer harvest browser cookies to bypass authentication entirely. Even when victims reset passwords and revoke sessions, attackers retain access through hidden device registrations — meaning standard incident response playbooks are now incomplete. Developers remain the preferred entry point for supply chain compromise. The Glassworm botnet was disrupted after hiding malware in VSCode extensions and npm packages, while the Megalodon campaign poisoned GitHub Actions workflows across 5,500 repositories. A malicious Sicoob.Sdk NuGet package stole banking certificates from Brazilian developers, and North Korea's Lazarus group compromised the widely used axios npm library — a single attack touching millions of downstream applications. Organizations must move beyond password-and-MFA reliance: adopt hardware security keys, shorten session lifetimes, delete attacker-registered devices before resetting credentials, and audit developer toolchains and CI/CD pipelines for tampering.

2026-05-285 minWithSecurehigh

WithSecure identified GREYVIBE, a Russia-nexus threat group targeting Ukrainian entities using spear-phishing, ClickFix, and fraudulent websites. The group systematically leverages Generative AI to develop custom malware (PhantomRelay, LegionRelay, FallSpy) and obfuscators, blending state-aligned intelligence gathering with cybercrime ecosystem overlaps.

Software Supply Chain and AI Exploitation Dominate Threat Landscape The software supply chain has become the primary battlefield for attackers because compromising a single developer tool can cascade into thousands of enterprise networks. Campaigns like Mini Shai-Hulud and TrapDoor are stealing credentials and injecting backdoors across major code registries, while the Laravel Lang Compromise and the Coruna Exploit Kit show how malicious code can automatically execute to steal secrets or exploit end users. As a result, organizations must treat developer environments as high-value targets, because a single compromised package or malicious VS Code extension can lead to catastrophic breaches like the GitHub internal repository theft by TeamPCP. In parallel, artificial intelligence is simultaneously accelerating attacks and creating dangerous new attack surfaces. Threat actors are using AI to automate influence campaigns like Patriot Bait and crack passwords, while also impersonating AI tools like Gemini CLI and Claude Code to deliver infostealers. Furthermore, attackers are directly targeting exposed AI infrastructure, such as Ollama AI endpoints, and manipulating AI coding assistants via hidden prompt injections in campaigns like TrapDoor, which means AI systems are both the weapon and the target. These trends together suggest that traditional perimeter defenses are failing against supply chain and AI-driven threats. Managers should immediately enforce strict vetting of open-source packages, restrict developer access to unverified extensions, and ensure AI infrastructure is not exposed to the public internet.

2026-05-247 minSocketcritical

The TrapDoor campaign is a sophisticated supply chain attack targeting crypto, DeFi, and AI developers across npm, PyPI, and Crates.io. The threat actor deployed over 34 malicious packages that utilize ecosystem-specific execution methods to steal credentials, wallets, and SSH keys, while uniquely leveraging AI configuration files like .cursorrules to trick AI assistants into executing exfiltration workflows.

2026-05-245 minCheck Pointcritical

This threat intelligence report highlights a surge in ransomware activity, critical zero-day vulnerabilities in Windows, and the active exploitation of Cisco Catalyst SD-WAN controllers. Additionally, it details emerging AI-driven threats, including malicious Hugging Face repositories and the abuse of AI website generators for phishing, alongside an APT intrusion by FamousSparrow targeting the energy sector.

2026-05-233 minZscaler ThreatLabzinfo

The article discusses the rapid enterprise adoption of agentic AI and emphasizes the need for deliberate innovation and governance. It highlights ACSC guidelines advocating for the integration of AI services into a Modern Defensible Architecture using principles like least privilege, segmentation, comprehensive logging, and human-in-the-loop oversight to mitigate the risks of autonomous compromise.

2026-05-234 minSocketmedium

The integration of AI coding tools has fundamentally altered the open-source ecosystem, driving an exponential increase in npm package creation and automating dependency selection. This shift transforms the software supply chain into an automated black box, bypassing traditional human review processes and elevating the risk of supply chain malware infections, thereby requiring automated behavioral analysis for mitigation.

2026-05-226 minAkamaihigh

A sophisticated attack campaign is targeting Ollama AI endpoints to deploy a custom Go-based P2P remote access Trojan (RAT) and cryptominer. The malware, named 'vc', leverages decentralized networking via libp2p to evade traditional C2 blocking and utilizes RAM disk execution and process masquerading to maintain stealth.

A financially motivated eCrime campaign is leveraging SEO poisoning to impersonate AI coding assistants like Gemini CLI and Claude Code, tricking developers into executing a fileless PowerShell infostealer. The malware executes entirely in memory, disables Windows telemetry (ETW and AMSI), and harvests sensitive enterprise credentials, session tokens, and files before exfiltrating them to attacker-controlled infrastructure.

2026-05-205 minPalo Alto Networkshigh

TamperedChef (also known as EvilAI) is a widespread threat campaign distributing trojanized productivity applications via malvertising. The threat actors heavily abuse legitimate code-signing certificates and employ delayed execution techniques to evade detection, ultimately deploying information stealers, RATs, or adware onto victim endpoints after a dormancy period.

2026-05-205 minAkamaihigh

Financial services are facing an escalating threat landscape characterized by massive DDoS attacks, AI-empowered botnets, and targeted web attacks against API endpoints. Attackers are increasingly exploiting overlooked DNS misconfigurations and leveraging hyperscale IoT botnets to bypass traditional IP reputation defenses, necessitating a shift toward behavioral heuristics and adaptive security architectures.

2026-05-197 minCisco Taloshigh

Cisco Talos has identified a commodity BadIIS malware ecosystem operating under a Malware-as-a-Service (MaaS) model, primarily used by Chinese-speaking threat actors for SEO fraud and traffic manipulation. The developer, known as 'lwxat', provides a dedicated builder and sophisticated service-based installers that ensure persistence on compromised Windows IIS servers while evading detection through custom Base64 encoding and service impersonation.

2026-05-194 minZscaler ThreatLabzinfo

The article highlights the critical need for foundational security architecture before deploying AI at scale, emphasizing that AI amplifies risks associated with exposed attack surfaces and lateral movement. It advocates for Zero Trust principles to make AI models invisible to the internet and restrict unauthorized access paths, preventing minor compromises from becoming systemic breaches.

Developer Supply Chains Under Siege as Edge Device Exploits Surge The dominant narrative this week is the coordinated weaponization of the software supply chain, as threat actors like TeamPCP and Mini Shai-Hulud aggressively target developer tools to steal cloud credentials. Because these attackers compromise trusted build systems like GitHub Actions, a single malicious package—such as the compromised TanStack libraries—can cascade into massive downstream breaches, allowing criminals to hold development environments hostage and even deploy destructive dead-man switches if their access is cut off. In parallel, attackers are bypassing traditional network defenses by exploiting internet-facing edge devices and logging in with stolen credentials. Threat clusters are actively exploiting critical flaws in Cisco Catalyst SD-WAN and Microsoft Exchange, while ransomware groups like The Gentlemen and state-sponsored actors like Secret Blizzard use these footholds to live off the land, hijacking legitimate IT tools to stay hidden for months. These trends together suggest that perimeter-focused defenses and basic patching are no longer sufficient. Organizations must immediately isolate their CI/CD pipelines from cloud credentials, enforce phishing-resistant multi-factor authentication on all internet-facing systems, and assume that trusted vendor tools may already be compromised.

2026-05-156 minAkamaicritical

The TeamPCP threat actor deployed the Mini Shai-Hulud worm in a sophisticated supply chain attack targeting the npm ecosystem via a GitHub Actions CI cache-poisoning technique. The malware steals credentials, establishes persistence via developer tools like VS Code and Claude Code, and features a destructive dead man switch that wipes the victim's home directory if access tokens are revoked.

2026-05-145 minCisco Talosmedium

The Talos Threat Source newsletter highlights an impending surge in software patching driven by AI vulnerability discovery tools. It also contrasts state-sponsored espionage tactics—which leverage valid credentials and native tools to bypass traditional defenses—with commodity ransomware, while summarizing recent supply chain compromises across developer platforms like Hugging Face and Jenkins.

2026-05-147 minCisco Taloscritical

Cisco Talos is tracking active exploitation of multiple vulnerabilities in Cisco Catalyst SD-WAN Controller and Manager. Threat actor UAT-8616 is exploiting CVE-2026-20182 for authentication bypass, while other clusters are chaining CVE-2026-20133, CVE-2026-20128, and CVE-2026-20122 to deploy JSP webshells and post-exploitation frameworks like Sliver and AdaptixC2.

Kimsuky (APT43) has updated its arsenal with new PebbleDash and AppleSeed malware variants, including the Rust-based HelloDoor and httpMalice backdoors. The group is increasingly utilizing legitimate services like VSCode Remote Tunnels, Cloudflare Quick Tunnels, and DWAgent for covert C2 and post-exploitation access, primarily targeting South Korean entities and global defense sectors.

2026-05-135 minAkamaihigh

Security researchers discovered critical vulnerabilities in three widely used Model Context Protocol (MCP) servers—Apache Doris, Apache Pinot, and Alibaba RDS—stemming from insufficient back-end security validation. These flaws include SQL injection (CVE-2025-66335), missing authentication, and unauthenticated data exposure, allowing attackers to execute arbitrary commands or exfiltrate sensitive database metadata.

2026-05-136 minSophoshigh

A malvertising campaign is leveraging a fake Claude AI website to distribute a malicious MSI installer. The infection chain employs DLL sideloading via a legitimate G DATA executable to execute DonutLoader, which ultimately deploys a novel backdoor dubbed 'Beagle' for remote command execution and file manipulation.

2026-05-134 minCofensehigh

Threat actors are increasingly leveraging Vercel's GenAI capabilities, specifically v0.dev, to rapidly generate and host highly convincing credential phishing pages. By combining AI-generated frontends with Telegram Bot API integrations for real-time credential exfiltration, attackers can deploy resilient, low-effort phishing infrastructure on legitimate cloud services that evades traditional detection mechanisms.

2026-05-134 minAkamaihigh

Model Context Protocol (MCP) servers introduce a new attack surface akin to AI-native APIs, exposing organizations to protocol-level attacks, injection vulnerabilities, and authorization bypasses. Because MCP tools often use permissive validation to accommodate LLM inputs and proactively broadcast their capabilities via plain-English descriptions, attackers can easily map business logic and exploit downstream systems or trigger resource exhaustion.

2026-05-136 minZscaler ThreatLabzhigh

Threat actors are exploiting the OpenClaw AI agent framework by publishing a deceptive 'DeepSeek-Claw' skill that distributes malware. The campaign utilizes malicious installation instructions to deploy Remcos RAT on Windows via DLL sideloading and GhostLoader on macOS/Linux via obfuscated Node.js scripts, enabling persistent access and data exfiltration.

2026-05-135 minRecorded Futurecritical

Recent research highlights severe security flaws in commercially available embodied AI systems, specifically Unitree humanoid and quadruped robots. Vulnerabilities including undocumented backdoors, hard-coded cryptographic keys, and unauthenticated APIs enable remote attackers to hijack devices, exfiltrate sensitive multimodal telemetry, and pivot across physical fleets via wireless interfaces.