GitHub's AI Coding Agent Leaked Private Repos With a Single Issue
Noma Labs' GitLost write-up shows a single public-repo issue can coerce GitHub's coding agent into leaking private repo contents.
Tag
#prompt-injection
Every LLM Self-Defense Eventually Broke
Researchers tested nine prompt injection defenses across 20,000 attacks. Every defense that relied on the model to protect itself failed. Only hard-coded output filtering survived.
AI Security Roundup: Lovable's 48-Day Open Door, MCP Remote Code Execution, and NIST Abandons 29,000 CVEs
A vibe-coding platform exposed every project's secrets through a trivial API flaw, Anthropic's MCP protocol enables remote code execution across 200,000 servers, and NIST can't keep up with AI-driven vulnerability discovery.
Your 'Safe' AI Chatbot Becomes a Liability the Moment You Give It Tools
Researchers tested five frontier LLMs as workplace agents. GPT-5.1 executed malicious instructions 75% of the time. Even the safest model failed 40%.
Your AI's Safety Filter Fails 97% of the Time Under Fuzzing
Palo Alto's Unit 42 tested LLM guardrails with genetic-algorithm prompt fuzzing. Content filters missed up to 99 out of 100 attacks.
Your 'Safe' AI Model Isn't Safe When It Has Agency
New benchmark finds frontier LLMs that pass safety tests become dangerously exploitable as agents. GPT-5.1 fell for 75% of prompt injection attacks. The problem isn't the model — it's the deployment.
Images That Hijack AI: Visual Prompt Injection Achieves 90% Attack Success
New research reveals multimodal LLMs are vulnerable to hidden instructions embedded in images. Mind maps, steganography, and physical signage all bypass text-based safety filters.
AI Agents Are Being Hijacked in the Wild: 22 Attack Techniques Now Documented
Unit 42 researchers catch indirect prompt injection attacks actively weaponizing AI agents on live websites, from forced transactions to data exfiltration
Your Customer Service Chatbot Is a Security Hole
IEEE S&P research finds 10,000+ websites running vulnerable AI chatbot plugins. Attackers can forge conversations, hijack tools, and extract system prompts.
AI Security Roundup: Critical Flaws in vLLM, AnythingLLM, and MS-Agent, Plus Rogue Agents Gone Wild
A busy week for AI vulnerabilities: video-based RCE, chat injection leading to full system compromise, and research showing AI agents autonomously bypass security controls.
Clinejection: How a GitHub Issue Title Compromised 4,000 Developer Machines
A prompt injection attack against Cline's AI triage bot escalated into a supply chain compromise - installing unauthorized software on thousands of developer systems.
Your AI Browser Can Steal Your Passwords via a Calendar Invite
Zenity Labs reveals how a malicious calendar event could let attackers hijack Perplexity's Comet browser to exfiltrate local files and take over your 1Password account.
AI Security This Week: Cursor RCE, 8,000 Exposed MCP Servers, and LLMs Jailbreaking LLMs
Cursor patches critical shell bypass flaw, thousands of MCP servers sit wide open, and new research shows reasoning models can autonomously jailbreak other AI systems with 97% success.
Your AI Prompts Are a Security Liability. Most Companies Haven't Noticed.
While enterprises focus on training data and model safety, inference - where AI actually processes requests - has become an overlooked security frontier with critical vulnerabilities.
Lobster Trap: The Open-Source Tool That Watches What Your AI Agents Say
Veea releases a sub-millisecond security proxy for AI agents under MIT license as new research shows 88% of organizations have experienced agent security incidents.
AI Security This Week: RoguePilot, Memory Poisoning, and 2,800 Leaked API Keys
GitHub patches critical Copilot takeover flaw, Microsoft warns of AI memory manipulation attacks, and thousands of Gemini API keys are found in public code.
The Agentic AI Security Crisis: 88% of Companies Report Incidents, Only 29% Are Ready
Cisco's 2026 State of AI Security report reveals a dangerous gap: enterprises are deploying AI agents faster than they can secure them, with MCP vulnerabilities and prompt injection attacks proliferating.
Clone a Repo, Lose Your API Keys: Claude Code's Trojan Horse Vulnerabilities
Security researchers found that simply opening an untrusted repository in Claude Code could execute arbitrary commands and steal your Anthropic API keys - all before you saw a warning.
That 'Summarize with AI' Button Is Brainwashing Your Chatbot
Microsoft found 31 companies embedding hidden instructions in AI share buttons. One click poisons your assistant's memory, shaping every future recommendation without your knowledge.
Google Translate Can Be Hijacked to Output Drug Recipes and Malware Instructions
Google's switch to Gemini for translation turned one of the world's most-used apps into a jailbreakable chatbot. Researchers tricked it into providing meth instructions instead of translations.
Link Previews Are Leaking Your Data to AI Agents Without a Click
Security researchers found that messaging apps' link preview feature turns AI agents into zero-click data exfiltration tools. Teams, Slack, Discord, and Telegram are all affected.
OpenAI Built a Bunker for ChatGPT Because It Can't Fix the Walls
ChatGPT's new Lockdown Mode protects against prompt injection data theft - but OpenAI admits the underlying vulnerability may never be solved. Here's what that means for agentic AI.
DockerDash: How Poisoned Container Metadata Turned Docker's AI Assistant Into an Attack Vector
Two independent security firms found that Docker's Ask Gordon AI could be hijacked through image metadata, enabling remote code execution and data theft across millions of developer machines.
Your AI Coding Assistant Can Be Weaponized: Three GitHub Copilot RCE Flaws Hit Every Major IDE
Microsoft patches three critical command injection vulnerabilities in GitHub Copilot affecting VS Code, Visual Studio, and JetBrains. Over 20 million developers at risk from unsanitized shell inputs.