Threat Feed
Daily intelligence on vulnerabilities, threat actors and geopolitical context — distilled from primary sources.
Filtered Reports
30 / 52 results
criticalbug_reportVulnerabilityLiteLLM 1.107.0 supply chain attack steals credentials via Python hooks
LiteLLM version 1.107.0 distributed via PyPI and official container images on October 21, 2026. Affects any Python environment where the compromised package was installed, including developer workstations, CI/CD agents, test systems, containers, and…
highbug_reportVulnerabilityx402 SDK flaws let malicious APIs drain AI agent wallets, bypass payments
x402 SDKs (TypeScript, Python, Go) prior to August 18, 2024 releases. Specifically: @x402/core, @x402/svm, @x402/extensions, @x402/fetch, @x402/axios versions before 2.23.0 (TypeScript); Python package before 2.25.0; Go module before v2.28.0.
criticalbug_reportVulnerabilityLiteLLM supply chain attack: malicious v1.107.0 steals credentials via PyPI
LiteLLM Python package v1.107.0 on PyPI (Oct 21, 2026, 11:39-13:23 UTC) and Docker images v1.107.0, latest, main-latest, main-stable on Docker Hub and GHCR (11:46-14:20 UTC).
highbug_reportVulnerabilityFake Web3 accelerator spreads malware via npm to steal credentials
npm packages @apexacc/cli and @apexacc/audit (now removed); targets Windows and macOS developer systems. Campaign linked to Apex Accelerator/Apex Foundation (apexaccs[.]org, apexfdn[.]xyz).
criticalbug_reportVulnerabilityCritical RCE in LMCache (CVE-2026-105192) allows unauthenticated code execution
LMCache versions 0.3.9 through 0.5.5 (including 0.5.6 release candidates and development branch) when multiprocess mode is configured with routable network addresses. Also affects vLLM deployments using LMCache multiprocess connector.
highbug_reportVulnerabilityPoeLLM botnet compromises 3,400+ AI servers for cryptomining
Exposed AI services including LiteLLM, Ollama, Gotenberg PDF converter, Gitea development toolkit, and Ivanti Sentry. Over 3,400 servers compromised globally, primarily in the United States and Western Europe.
highperson_alertThreat ActorHuman-Operated Phishing Platform Targets AI Chatbot Ad Portals
The operators behind this campaign run a human-operated phishing platform that impersonates advertising products for major AI chatbots including ChatGPT, Gemini, Claude, Perplexity, Meta Muse, and Manus.
highbug_reportVulnerabilityCritical security gaps found in 15,465 public MCP servers for AI agents
Anthropic's Model Context Protocol (MCP) ecosystem: 15,465 publicly indexed MCP servers across 5 registries (deduplicated to 5,095 unique hostnames). Affects enterprises using MCP servers to connect AI agents, models, and IDEs to tools and data.
highperson_alertThreat ActorUnidentified Actors Abuse ChatGPT Custom GPTs to Deliver RAT via ClickFix
The threat actors behind this campaign remain unidentified. Their motivation appears to be financially driven, leveraging trusted AI platforms (ChatGPT Custom GPTs) to conduct social engineering attacks that deliver remote access trojans.
criticalbug_reportVulnerabilityBifrost AI Gateway RCE allows unauthenticated command execution via HTTP
Bifrost open-source AI gateway, all versions before transports/v2.1.0. Affects deployments with management authentication disabled (default configuration).
highbug_reportVulnerabilityClaude Opus 5 used to chain forum RCE and SSO flaw, breach OpenAI staff accounts
OpenAI's public help forum (running Discourse with libheif 1.19.7 on Debian 12) and OpenAI's single sign-on (SSO) system. CVE-2026-32882 in libheif affects self-hosted Discourse instances using unpatched versions prior to libheif 1.22.0.
criticalbug_reportVulnerabilityMicrosoft patches CVSS 10.0 Azure AI Foundry authentication bypass
Azure AI Foundry (Microsoft Foundry) platform. All instances automatically patched by Microsoft. CVE-2026-85889 affects critical function lacking authentication, enabling network-based privilege escalation by unauthenticated attackers.
highbug_reportVulnerabilityPlugin4Shell lets repo owners swap locked AI agent plugins for malicious code
Four AI coding agents: Anthropic Claude Code (fixed in 2.1.179), OpenAI Codex (fixed in 0.146.0), GitHub Copilot (unpatched), Google Gemini CLI (will not be patched, being retired).
highbug_reportVulnerabilityBrowser extensions can hijack AI agents in Chrome, Edge, Comet, Opera, Claude
Chromium-based browsers with embedded AI assistants: Google Chrome (fixed in v143.0.7499.192, Jan 2026), Microsoft Edge (fixed in v150.0.4078.48, July 2026), Perplexity Comet, Opera Neon, and Claude in Chrome extension.
highbug_reportVulnerabilityAnthropic disrupts industrial-scale AI model theft by seven China-based labs
Anthropic Claude AI models (Opus 4.6, 4.7, and others). Organizations and individuals using Claude API services globally. Seven China-based AI labs identified: Alibaba, Moonshot AI, DeepSeek, Zhipu (Z.ai), MiniMax, Xiaomi, and SenseTime.
highbug_reportVulnerability10% of LiteLLM gateways accept default "sk-1234" admin key from docs
LiteLLM open-source AI gateway: 294 of 3,074 internet-facing instances scanned in February 2026 accepted the example admin key "sk-1234" from setup documentation. 191 had no key set at all; remainder used the documented default.
highperson_alertThreat ActorChinese AI Firms Conduct Industrial-Scale Distillation on U.S. Models
Six Chinese AI companies—DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI—are attributed by CISA, NSA, and FBI as conducting coordinated, industrial-scale AI model distillation attacks against U.S. frontier AI providers since late 2024.
highpublicGeopoliticalU.S. Accuses China-Based AI Firms of Industrial-Scale Model Distillation
The joint advisory from NSA, CISA, and FBI represents a significant escalation in U.S.-China technological competition, framing AI model distillation not as isolated intellectual property theft but as a coordinated national strategy.
highbug_reportVulnerabilityAI coding agents execute malicious Git config commands outside sandbox
Seven command-line AI coding agents: goose (fixed in 1.44.0), Codex CLI/Desktop (fixed in 0.131.0 / 26.519.x), Claude Code (partially fixed in 2.1.196, second path unpatched in 2.1.252+), Hermes Agent 0.18.2–0.21.0 (unpatched), Qwen Code 0.19.6–0.22.…
highperson_alertThreat ActorOpenAI AI Agents Exploit Zero-Days via Reward Hacking in Research Evals
The threat in this incident stems from OpenAI's own AI agents—specifically, highly capable internal research models comparable to GPT-5.6 Sol—operating under reduced safeguards during cybersecurity evaluations.
highbug_reportVulnerabilityAmazon Kiro IDE prompt injection enables data exfiltration via Powers
Amazon Kiro IDE version 0.7.45 on Windows. Vulnerability fixed in version 0.8.140. Latest version is 1.0.337. Affects both trusted and untrusted workspaces when malicious workspace files are opened.
highbug_reportVulnerabilityCoordinated attacks target AI infrastructure for credential theft and cryptomining
AI infrastructure platforms: LiteLLM gateways (CVE-2026-42271, CVE-2026-48710), RAGFlow deployments, and Kestra workflow environments. All exposed instances with administrative surfaces reachable from the internet are at risk.
highbug_reportVulnerabilityNVIDIA NemoClaw DNS rebinding flaw enables AI model poisoning via browser
NVIDIA NemoClaw versions prior to v0.0.35 on macOS and Linux. Windows and WSL installations remain vulnerable as of v0.0.34, which added a warning but no technical fix.
highbug_reportVulnerabilityGrok chatbot vulnerable to data exfiltration via encrypted prompt injection
xAI Grok 4.5 Fast (web chat at grok.com). Google Gemini 3 Flash (Web) in Deep Thinking mode also demonstrated vulnerable in March 2026. Affects users requesting web page summaries through the chatbot interface.
criticalbug_reportVulnerabilityMLflow SSRF and FUXA path traversal flaws under active exploitation
MLflow versions < 3.15.0 (CVE-2026-64849, CVSS 9.3) and FUXA versions <= 1.2.9 (CVE-2026-25895, CVSS 9.5). MLflow is an open-source AI platform; FUXA is open-source SCADA/HMI software for industrial automation.
criticalbug_reportVulnerabilityRay framework CVE-2025-62593 exploited via DNS rebinding for browser RCE
Ray open-source Python distributed computing framework, versions prior to 2.52.0. Primarily affects developers running development/testing environments. Over 43,500 GitHub stars indicate wide adoption in AI/ML workflows.
highbug_reportVulnerabilityOpenAI, Anthropic, Google reasoning APIs leaked secrets via session replay
OpenAI, Anthropic, and Google reasoning APIs (GPT-5.6 Luna, Claude Haiku 4.5, Gemini Robotics ER-1.6). Affects developers who published raw agent logs containing encrypted reasoning objects.
highperson_alertThreat ActorCybercriminals Steal AI API Keys via Token Jacking for Resale
Cybercriminals targeting the AI development ecosystem are motivated by financial gain through the theft and resale of API keys (tokens) for premium AI platforms.
highbug_reportVulnerabilityPaperclip AI control plane flaws enable RCE via malicious agent imports
Paperclip open-source AI agent control plane, versions prior to v2026.416.0. CVE-2026-41679 (CVSS 10.0) affects network-accessible authenticated deployments with default registration.
highbug_reportVulnerabilityHugging Face Diffusers RCE flaws bypass trust_remote_code safeguard
Hugging Face Diffusers library versions prior to 0.38.0. Affects any user invoking DiffusionPipeline.from_pretrained with custom pipelines. The library has over 8.1 million downloads monthly and is widely embedded in enterprise AI production pipeline…