Malicious GitHub Repos Turn AI Assistant Configs Into Attack Paths
Attackers are using AI coding setup files as an execution route. Bitdefender Labs has identified 22 malicious GitHub links that try to steer...

Created by weiqun zou
Curated AI coding assistant incidents with source links, code snippets, and detailed timelines
Explore the latest content tracked by AI Coding Incident Tracker
Attackers are using AI coding setup files as an execution route. Bitdefender Labs has identified 22 malicious GitHub links that try to steer...
The core critique is that per-change checks run through skill files and a Chrome tool are expensive, and Playwright could reach equal quality...
A Palo Alto Networks survey found enterprises let AI agents change code and production environments, yet they can't trace those agents or quickly stop...
Cursor says the tampered extension traces back to a leaked key issued to partner Fireworks.
A Palo Alto Networks report, based on a survey of 2,800 security executives and practitioners across 10 countries, points to an operational security...
California Attorney General Rob Bonta is suing OpenAI, accusing it of knowingly testing advanced models without adequate security and misleading the...
The article argues coding agents choose and install packages mid-task, so the usual validation window never opens. A package's install script runs...
The piece opens with a blunt premise: "Your AI Agent Says It Fixed the Bug. Make It Prove It."
It starts from the core problem: some bugs only show...
Detectify's new Agentic Tooling lets an organization's own AI agents test whether suspected vulnerabilities can be exploited in running...
A hobbyist says Claude did most of the diagnosis on his 1994 Jeep Cherokee's computer, while Codex handled the memory work. The link ran at about...
Lauren Tan says agents at Cursor loved setTimeout quick fixes and justified them with comments. In a separate count of 8,118 VS Code commits,...
According to Cursor's November 13 update, the attacker started with an access key Cursor issued to partner Fireworks. Unredacted credentials in...
This piece is general threat analysis, not a documented case of an AI agent causing damage. Its core argument is that developer machines hold source...
ArmourCode has released an open-source tool, Claude Code Hooks, that adds pre-execution checks to block certain commands before they run. It is built...
Cursor's own investigation traces the attack to a third-party key, but the impact figures are the company's claims.
Sonar's developer survey puts numbers on the verification gap. 76% say they generate AI code faster than they can verify it, and 53% now spend more...
Read-only access doesn't close the risk. Query results go into the model's context, so any customer data the agent selects leaves your database....
According to one report, Anthropic removed a Claude Code feature on November 13, 2026. Developers had reportedly reached it through an undocumented...
Researchers say hijacked GitHub accounts were used to open probe pull requests across 61,764 repositories, starting Nov. 10, according to Aikido...
Railway's postmortem separates a worst-case ceiling from the figures it has reported: