Library · security

AI agent security: the complete map

Added 2026-08-16 · Last verified 2026-08-16

Pillar hub mapping the four agent attack surfaces — input (prompt injection), memory (poisoning), execution (sandboxing), authority (guardrails/approval gates) — with links to the deep-dive guide for each, plus real incident case studies (Langflow KEV, Therac-25, Knight Capital).

Short answer: securing an AI agent means defending four distinct attack surfaces — what goes into the model (injection), what the model remembers (memory), what the model runs (code execution), and what the model is allowed to do (authority). This hub maps each surface to a deep-dive guide.

1. Input: prompt injection

Any text your agent reads — web pages, emails, tool output — can carry instructions aimed at the model. Injection remains the #1 practical agent exploit because it needs no access to your infrastructure, only to something your agent will eventually read. → Prompt injection defense

2. Memory: poisoning the well

Persistent memory turns a one-shot injection into a standing compromise: plant a fact once and the agent re-reads it forever. Memory needs provenance, quarantine for un-trusted writes, and periodic audits. → Defending against memory poisoning

3. Execution: sandboxing generated code

Agents that write and run code need real isolation — microVMs or user-space kernels, locked-down egress, scoped credentials — because LLM-generated code must be treated as hostile by default. → Sandboxing agent code execution

4. Authority: guardrails and approval gates

The blast radius of a compromised agent equals the permissions you gave it. Cap it with explicit guardrails, human approval for irreversible actions, and least-privilege tool scopes. → Guardrails and safety · Human approval gates

Case studies

Real incidents beat theory. The Langflow RCE reached CISA's KEV list because an agent-builder exposed unauthenticated code execution: → Langflow CVE in CISA KEV

And the pre-LLM history of safety-critical software failure still sets the frame for what "unsafe autonomy" costs: → Therac-25 · Knight Capital

The one-paragraph security model

Treat the model as an enthusiastic intern with no loyalty: everything it reads may be adversarial (injection), everything it remembers may be planted (poisoning), everything it executes may be a payload (sandboxing), and everything it's permitted to do will eventually be done wrong (authority). Defense-in-depth across all four layers — not any single fix — is what makes an agent deployable.

Useful?

Read the source ↗


Source: The Agent Loop — agent-loop.xyz. Curated in the open knowledge base (CC BY 4.0).

Ask a question about this ↗