A guardrail is only a guardrail if it can stop something. Three layers do the work: what you prevent before the agent starts, what you enforce while it runs, and what you detect and correct after. Every control below is executable, and each one is marked hard or soft, because the difference decides what you can rely on.
Hard · the boundary
Soft · the signal
Controls that behave identically whatever the model produces, because the model is not what evaluates them.
Valuable, worth building, and not the thing standing between an agent and an irreversible action.
| Risk | What it looks like | Primary control | Where it lives |
|---|---|---|---|
| ASI01 Goal hijack | Injected text in a document or web page redirects the agent to a different objective | Fixed tool set and envelope for the task; retrieved content in a labelled data channel; task contract with stop conditions. Screening supplements. | R2 Gateway |
| ASI02 Tool misuse | A legitimate tool is called with parameters that were never intended, or chained into an unintended effect | Per-call authorization, parameter allow-lists, value limits, sequence monitoring | R2 · L6 |
| ASI03 Identity and privilege abuse | The agent acts with more authority than the user who asked, or keeps access after the user leaves | Delegated scope only, short-lived tokens, orphan detection, access re-certification | Governance rail |
| ASI04 Supply chain compromise | A malicious or hijacked MCP server, skill or prompt pack enters the runtime | Signed and pinned artifacts, private registry, scanning, provenance checks | L6 Registry |
| ASI05 Unexpected code execution | Generated code runs with real credentials or reaches the network unexpectedly | Ephemeral sandbox, egress deny-list, no standing credentials, output code review | R4 Runtime |
| ASI06 Memory and context poisoning | False facts written to durable memory persist and are treated as trusted later | Provenance required on every memory entry, scoped writes, TTL, correction path. Poisoning evals measure whether it holds. | R4 Memory |
| ASI07 Insecure inter-agent communication | Agents trust each other's messages without verifying who sent them or on whose authority | Mutual authentication, signed A2A messages, propagated user scope, peer allow-list | R3 Orchestration |
| ASI08 Cascading failures | One bad output becomes another agent's trusted input and the error multiplies across a chain | Circuit breakers, step and delegation-depth budgets, blast-radius caps, confidence thresholds. Guardian monitoring detects; it does not bound. | R3 Orchestration |
| ASI09 Human-agent trust exploitation | Confident output plus approval fatigue means a person rubber-stamps something harmful | Show the diff and the evidence rather than the summary. Sample-audit approvals. Rotate approvers. | Human control |
| ASI10 Rogue agents | An unregistered, forgotten or compromised agent operates outside all of the above | Registry as the allow-list, network egress control, discovery scanning, automatic expiry | Evidence rail |
If you are standing up an agent platform, tightening the controls on one you already have, or preparing for an audit that now includes agents, I am happy to look at it with you.
Tell me where you are with agents and what you are trying to make safe. I reply to every enquiry within two business days.