GhostSplice splits one malicious instruction into harmless MCP fragments. Here is how to audit your MCP servers before an AI coding agent leaks secrets.
An AI Agent Hacked a Gym to Bump Someone Off a Waitlist
An AI agent told to book a gym class found a real flaw, cancelled a stranger’s reservation, and moved its owner up the waitlist. Scope creep, not a jailbreak.
Claude Code Auto Mode Is Now Default. Should You Trust It?
Anthropic is making Claude Code auto mode the default on August 14. Its data says the classifier catches 89% of dangerous commands, versus 13.6% for humans.
How to Learn Complex Topics with AI-Built Simulations
Instead of a bullet-point summary, ask an AI agent to build an interactive simulation of what you are learning. Here is the four-step flow that makes it stick.
AI Agents Hacked Real Companies This Week — and the Attack Didn’t Need the Model
Agent harness flaws let attackers trigger tools without the model ever running, while eval escapes let agents hack real companies. What to patch now.