This popular AI agent could be hacked by a single email — with potentially disastrous consequences

Researchers found a way around Manus' guardrails and got it to execute a simple email prompt injection attack.
SC Media

Researchers bypass AI agent protections with JavaScript obfuscation

Security researchers have successfully bypassed the prompt-injection protections of an AI agent named Manus, achieving code ...
The Hacker News

Tensorlake npm Package Compromised to Deliver Shai-Hulud Credential-Stealing Worm

Malicious tensorlake npm version 0.5.144 contained a Shai-Hulud worm that harvests credentials and can republish compromised ...
Cyber Security News

GlassWorm Supply Chain Attack Uses Fake VS Code Themes to Deliver Hidden Malware

GlassWorm hides malware in VS Code theme extensions, targeting developers through Visual Studio Marketplace and Open VSX.

[Explanation] Beware of North Korean Fake Companies: Cyberattacks Targeting Engineer Job Hunting

Police, judicial authorities, and security agencies in Japan, the U.S., and other countries have issued urgent warnings ...
Security Boulevard

How We Hijacked an AI Agent With a Single Email

Executive SummarySalt Labs found that the agentic AI platform Manus could be hijacked with a single email. By hiding ...
cyberpress

Malicious VS Code Themes Hide GlassWorm Malware Targeting Developers

A group of VS Code theme extensions linked to GlassWorm, a malware campaign targeting developers. Their investigation ...

Malware Hiding in Claude Code's Configuration Folder .claude/ — Reading Google's Threat Report

A report published by Google's Threat Intelligence Group (GTIG) on September 9, 2026, details MCP server hijacking and supply ...
MarkTechPost

What Happens When a Trusted Model Repo Changes? Unsloth Studio Re-Checks Before It Runs

Unsloth details how Studio scans model code, blocks flagged weights, inspects packages and sandboxes tools before anything ...