Key figure: 63% of organizations in IBM's 2025 breach study had no AI governance policy, and 97% of AI-breach victims had no proper AI access controls (IBM, Cost of a Data Breach Report 2025).
Every AI agent runs with a set of permissions — files it can read, tools it can call, money it can spend. Most permission sets are far broader than the job requires, which is exactly how agents get hijacked into doing an attacker's bidding. This quiz shows five realistic permission sets; you vote HIRE or FIRE, then see what each one actually allows.
Production database write access, sending email or messages as you, code execution on your machine, and spending money — none belong in a default permission set. The principle is least privilege: an agent gets the minimum access its task requires, and anything destructive needs human approval. Every agent hijack exploited permissions broader than necessary, turning a trusted assistant into an attacker's insider.
Least privilege is the security principle that every user, program, and AI agent should have only the minimum access needed to do its job — nothing more. It limits the damage any single compromise can cause.
5 rounds. Each one shows you a job description that is actually an AI agent's real tool-permission set. Vote HIRE or FIRE — then we'll show you what the agent could genuinely do with those permissions, and the AI agent security best practice that fixes it.
⚠️ Illustrative scenarios. These are realistic permission sets assembled for education, not reviews of any real product. No specific agent, vendor, or app is being described or graded.
Agent Auth Gateway gives every AI agent a budget, an allowlist, an approval queue, and a kill switch — so the permissions you grant are the permissions they actually use.
Get Agent Auth Gateway Early Access →Educational content. The reveals describe plausible capabilities of the illustrated permission sets — factual about what those permissions enable, not predictions of what any real agent will do. Guardrails shown are standard least-privilege practices. Built by Ingenious AI.
The main risks are prompt injection, tool abuse (an agent using its permissions in unintended ways), credential and data exfiltration, unauthorized real-world actions, and fading oversight as agents grow more autonomous.
Use least-privilege permissions, require human approval for irreversible actions, enforce egress allowlists at the network layer rather than in the prompt, keep tamper-evident logs of every agent action, and maintain a working kill switch.
Least privilege means an agent gets only the permissions its task strictly needs — no standing admin, no broad data access, no irreversible actions without approval. The quiz shows what over-broad permission sets actually let an agent do.