Sign InOpen Brain
AI EngineerVideoSource Linked

IT Admin for the AI Workforce — Sarthak Aggarwal, Decawork

Agent instructions are not security boundaries. Put identity, scoped capabilities, policy gates, audit receipts, and revocation outside the model before granting production access.

AI Engineer · Aug 20, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Agents that read untrusted content while holding private data and action rights can turn text into side effects. The talk contrasts **EchoLeak**, a zero-click Microsoft 365 Copilot chain, with the **Replit production-data incident**, where no attacker was required.

Practical Implication

Treat every agent as a runtime identity acting for a named subject. Convert authenticated intent into a typed plan, give the executor short-lived capabilities, and place a deterministic policy gate before each tool call. Keep destructive approvals and revocation outside the model.

Agent-Ready Context
Agents that read untrusted content while holding private data and action rights can turn text into side effects. The talk contrasts **EchoLeak**, a zero-click Microsoft 365 Copilot chain, with the **Replit production-data incident**, where no attacker was required.

Treat every agent as a runtime identity acting for a named subject. Convert authenticated intent into a typed plan, give the executor short-lived capabilities, and place a deterministic policy gate before each tool call. Keep destructive approvals and revocation outside the model.

Privilege separation reduces authority; it does not make hostile context or model errors disappear. MCP and A2A provide communication rails, but the talk argues they do not supply ownership, delegation policy, or a complete audit trail.
Connected Context · Feed7 Judgment

This elevates agent security from sandboxing alone to identity and authority design: bind actions to a named subject, translate intent into typed plans, issue short-lived capabilities, and enforce deterministic policy before every tool call. It confirms runtime separation patterns while clarifying that MCP or A2A connectivity does not establish ownership, delegation, revocation, or auditability.

In Code They Act, In Proof We Trust — Erik Meijer, Leibniz LabsBoth separate planning from side effects and require a machine-enforced gate before execution; this talk adds subject identity, short-lived delegation, and revocation requirements.Unlock Agent Autonomy: The Runtime for AI-Native Systems — Tushar Jain, DockerDocker’s external capability approval and task containment implement the same principle that authority decisions must remain outside the agent’s trust boundary.zhaoxuya520/reverse-skillThe security router’s authorization-first planning and evidence trail align with typed intent and auditing, while this talk demands stronger runtime identity and per-call capability enforcement.Stop AI Agent Hallucinations: 5 Techniques + Production Patterns - Elizabeth Fuentes, AWSPre-call rule enforcement supports the deterministic policy-gate requirement, while typed plans and delegated identities extend it beyond response validation and runtime correction.
Context Map
agentcodingsecurity#harness-engineering#tool-use#sandboxing
Uncertainty
Privilege separation reduces authority; it does not make hostile context or model errors disappear. MCP and A2A provide communication rails, but the talk argues they do not supply ownership, delegation policy, or a complete audit trail.