ThreatNeuron
AI security. Cybersecurity. The stuff that matters.
Recent
Shadow AI: The Hidden Threat Leaking Enterprise Data
·2029 words·10 mins
Unauthorized AI tools are the fastest-growing blind spot in enterprise security. With up to 80% of employees using unsanctioned AI at work, shadow AI is leaking credentials, customer data, and proprietary code through channels most security teams can’t even see.
React2Shell CVE-2025-55182: Mass Harvesting AI API Keys
·1789 words·9 mins
A CVSS 10.0 vulnerability in React Server Components is being actively exploited to steal AI platform API keys, cloud credentials, and payment secrets from hundreds of Next.js deployments. Here’s how the attack works and what your team should do right now.
GrafanaGhost: Prompt Injection Turns Dashboards Into Spy Tools
·1791 words·9 mins
Noma Security’s GrafanaGhost disclosure reveals how indirect prompt injection in Grafana’s AI assistant can silently steal financial metrics, infrastructure data, and customer information — without authentication. Here’s what happened, how it works, and what to do about it.
CVE-2026-39987: Marimo RCE Flaw Exploited Within Hours
·1623 words·8 mins
A missing authentication check in Marimo’s terminal WebSocket endpoint handed attackers root shells on AI notebooks across cloud platforms. Exploitation began less than 10 hours after disclosure, with credential theft completed in under 3 minutes per target.
Project Glasswing: Anthropic's Mythos AI Finds Thousands of Zero-Days
·1867 words·9 mins
Anthropic’s Project Glasswing pairs the unreleased Claude Mythos AI model with 50+ partner organizations to find and fix thousands of high-severity zero-day vulnerabilities in critical software. The initiative has already uncovered bugs hiding in codebases for decades.
Flowise CVE-2025-59528: CVSS 10.0 RCE Flaw Hits 15K AI Platforms
·1706 words·9 mins
A maximum-severity RCE flaw in Flowise’s CustomMCP node is under active exploitation, putting thousands of AI workflow platforms at risk. Here’s what security teams need to know about CVE-2025-59528 and how to respond.
OpenClaw Security Crisis: 135K Exposed AI Agents Under Siege
·1769 words·9 mins
Over 40,000 OpenClaw AI agent instances were found exposed to the internet, with 63% vulnerable to remote exploitation. Combined with a coordinated supply chain attack on the ClawHub marketplace that planted 824 malicious skills, the OpenClaw crisis is a wake-up call for anyone deploying agentic AI.
AI Cyberattacks 2026: How Attackers Weaponize Every Stage
·1668 words·8 mins
Threat actors are embedding AI into every phase of the cyberattack lifecycle — from reconnaissance to data exfiltration. This post breaks down how platforms like Tycoon2FA exploit AI for phishing at scale, what the weaponized attack lifecycle actually looks like, and where defenders should focus their resources.
AI Device Code Phishing: Bypassing MFA at Scale
·1792 words·9 mins
Device code phishing has become the go-to technique for AI-powered campaigns that sidestep multi-factor authentication entirely. This post breaks down how the attack works, why AI makes it dramatically more effective, and what defenders can do right now.
LiteLLM Supply Chain Attack: 40 Minutes That Hit 500K Systems
·1747 words·9 mins
A supply chain attack on LiteLLM pushed malicious packages to PyPI that harvested credentials from an estimated 500,000 machines in under an hour. This post dissects the attack chain, the cascading damage across AI infrastructure, and the hard lessons for organizations running open-source AI tooling.
AI Agent Hijacking: Weaponizing Autonomous Systems
·1943 words·10 mins
Autonomous AI agents are becoming prime targets for attackers. This post breaks down the techniques behind AI agent hijacking, real-world attack chains, and what security teams can do to defend their agentic infrastructure.
Understanding Prompt Injection Attacks: A Comprehensive Guide
·483 words·3 mins
Prompt injection is one of the most significant security risks facing AI-powered applications. This guide breaks down how these attacks work and what you can do about them.