Librería Samer Atenea
Kálamo Books
Librería Elías (Asturias)
Librería Kolima (Madrid)
Librería Proteo (Málaga)
Give an AI agent access to your customer database, email, and billing system, and it can get real work done. But what happens when it starts taking orders from someone else?The danger isn’t always a hacker breaking through your defenses. Sometimes it’s a perfectly authorized AI following instructions it should never have trusted. As businesses rush to connect artificial intelligence to their systems, a fundamental question remains: who actually controls what the AI is allowed to do?In Taking Orders From Strangers, Will Holmes puts four approaches to AI tool security to the test, including MCP, UTCP, and two new security architectures developed through his research, OTCP and LTCP. Through more than a thousand experiments and fifteen categories of attack, he examines prompt injection, access control, and the limits of trusting AI with real authority.The findings challenge a common assumption: protecting a business isn’t simply about making AI more trustworthy. It’s about building systems that remain secure when the AI gets something wrong.