This page may contain stale information. Last updated: 2026-06-12
Definition
The Authorized Intent Chain (AIC) is a security concept coined by Tenet Security (June 2026) describing attacks where every step in the kill chain is individually authorized — causing traditional security controls (EDR, firewalls, IAM, VPNs) to see no unauthorized behavior.
Key Points
- Root cause: Security models detect unauthorized actions; AI agent workflows make attacker-injected data appear as legitimate user intent
- Mechanism: Developer asks agent to fix errors → agent queries trusted MCP tool → injected data becomes “authorized” remediation → agent executes with developer privileges
- No malware signature: No phishing, no credential theft, no network anomaly — purely data injection into trusted channels
- Prompt injection immunity: System prompts instructing agents to ignore untrusted data failed in agentjacking tests
Contrast with Traditional Attacks
| Dimension | Traditional RCE | Authorized Intent Chain |
|---|---|---|
| Entry vector | Exploit/vulnerability | Injected data in trusted tool |
| EDR detection | Malicious process/network | All actions authorized |
| User interaction | Often requires click/phish | Developer’s normal workflow |
| Fix location | Patch vulnerable service | Agent runtime execution layer |
Implications for AI Agent Security
- Perimeter and endpoint defenses insufficient for agentic workflows
- MCP integrations returning externally-influenced data create systemic risk beyond any single vendor
- Defense must shift to execution-layer controls: intercept tool calls, enforce least-privilege, require human approval for high-impact actions