AgentTag vs shared API keys — why AI agents need their own identity
Most AI agent deployments give agents shared human API keys or environment-variable credentials. AgentTag replaces this pattern with per-agent cryptographic identities, signed mandates, and tamper-evident audit trails — giving teams scope, attribution, and clean revocation that shared credentials cannot provide.
At a glance comparison
| Security Dimension | Shared API Keys | Agent Passport |
|---|---|---|
| Identity Type | Shared (impersonates human credentials) | Cryptographic (Ed25519 Keypair + W3C DID per agent) |
| Attribution & Auditing | Indistinguishable (logs register human operator) | Attributable (every action signed by specific agent) |
| Permission Scope | Broad/Unscoped (accesses everything key allows) | Least Privilege (tool, parameter, spend, and time caps) |
| Revocation Process | Collateral (requires rotating key used by others) | Targeted (revoke single mandate in real time) |
| Blast Radius | Unlimited | Bounded by active mandate scope |
| Prompt Injection Defence | None (executes whatever model requests) | Policy Gate (blocks out-of-mandate operations) |
The three failure modes of shared credentials
1. The Scope Vulnerability
When an agent uses a shared API key, it has broad, unchecked permissions. An autonomous coding agent compromised via malicious repository code could delete files, download secrets, or purchase paid SaaS upgrades because the shared credential lacks parameter-level boundaries.
2. The Attribution Void
Audit logs from your cloud providers and tools will record all actions as having been performed by the account owner. You cannot separate actions initiated by human operators from actions run by the LLM agent — rendering incident investigation impossible.
3. The Cascade Revocation Problem
If a single agent's access token is leaked or its execution logic is hijacked, disabling the access key breaks every other agent and automated workflow sharing that token. The resulting downtime causes teams to delay revoking access, compounding the breach duration.
Frequently asked questions
Why is giving an agent my API key unsafe?
+
It gives the agent broad, unscoped access. If the agent misbehaves due to model error or prompt injection, it has full permissions to execute any commands the key allows. It also makes auditing impossible, as all actions show up in logs as if they were done by you.
What is an agent passport?
+
An agent passport is a cryptographic identity (Ed25519 keypair bound to a W3C DID) issued specifically to one agent. Every action is signed using this key, providing verifiable attribution and isolation.
How does AgentTag enforce scope constraints?
+
AgentTag enforces constraints using signed mandates. A mandate dictates what tools, parameters, spending thresholds, and expiry dates apply to the agent, checking every request in real time.
Is mandate revocation instant?
+
Yes. Since mandates are evaluated at request time, revoking a mandate in the control plane invalidates permissions instantly. You do not need to restart the agent or rotate keys.
What is the performance latency with AgentTag?
+
AgentTag's policy engine checks mandates with a latency of less than 50 milliseconds, adding virtually zero noticeable overhead to model response times.
Does AgentTag see our raw application data?
+
No. AgentTag only processes and logs execution metadata (timestamps, tool names, mandate IDs, status codes). We do not record or view the raw contents of requests or responses.
Switch from API keys to agent passports
Protect your production integrations using cryptographic identity and mandates. Free during public beta.