Start a 90-day pilot

Glossary

Ten terms a security team meets once its agents hold tools. Each is defined here in a line and in full on its own page. Under every term is what Intercis does about it today, and where it does not do the thing the term describes, that line says so. Longer arguments are in the blog; the numbers are on the validation report.

One boundary governs all ten. Intercis judges the tool calls that cross the Anthropic and OpenAI API wire. A tool a hosted MCP server runs for the model has already run by the time the response reaches the proxy, so it is observed, not governed: one observe row, never judged or blocked.

  1. Agent Identity Validation

    Enforced when the key is bound to one agent.

    Checking that an agent is registered, authorized, and working inside its declared tool scope and rate limits. OWASP T9. Bind an API key to one agent and possession of the key is the identity; leave it unbound and the agent header is self-asserted, so any holder of the key can claim any agent of that tenant.

  2. Agentic AI Threats

    We use its numbers. We do not claim coverage.

    The OWASP taxonomy of threat categories for agents with tool access, T1 to T17 in the v1.1 edition. We publish no coverage score against the list.

  3. AI Agent Audit Trail

    One row per governed call, hash-chained per tenant.

    A tamper-evident, append-only record of what an agent did: the policy decision, the timestamp and the threat context. It is not tamper-proof.

  4. AI Agent Governance

    Enforcement and the record only.

    Deciding, before a tool call runs, whether the agent may make it, and keeping the record. Intercis does not run the human approval step. The longer guide to AI agent governance is on the blog.

  5. AI Agent Kill Switch

    Manual, one agent at a time.

    A stop that takes an agent's access away. Ours is manual and per-agent: an operator deactivates the agent and the proxy denies its next call with a 403. There is no anomaly score behind it and no automatic session kill anywhere in the product.

  6. AI Agent Policy Enforcement

    Allow, deny and observe, on the calls it sees.

    Allow, deny and observe decisions applied to an agent's actions by a deny list of compiled patterns. Those are the three live verdicts; the audit table's verdict column allows a fourth, pending, for a call that would need human review, and nothing writes an approval against it yet.

  7. AI Agent Proxy

    Intercis is one.

    An intercept proxy between agent code and the model API. It reads tool calls crossing the Anthropic and OpenAI wire and enforces policy before they execute.

  8. Human-in-the-Loop

    Not run here.

    An approval step that asks a person to authorize a high-risk action before the agent takes it. Intercis does not run one. The audit table keeps pending as a verdict for a call that would need review, and no reviewer approval loop is wired up to it.

  9. Prompt Injection

    Detected and logged, never blocked.

    An attack that rewrites an agent's instructions to point it at actions nobody authorized. OWASP T6. Intercis detects and logs it; it does not block on it.

  10. Tool Call Interception

    The Anthropic and OpenAI API wire only.

    Inspecting the tool calls an agent makes over the Anthropic and OpenAI API wire, and enforcing policy on them before they execute.

Intercis is a proxy that runs outside your agent process. Your agent's traffic to Anthropic or OpenAI goes through it, policy runs on the tool calls the model asks for, and once enforcement is on, a denied call is taken out of the response before your agent runtime can execute it.

Start a 90-day pilot

Every line here that says what the product does was read against the code on , and each entry page carries that date too.

The T-numbers come from Agentic AI Threats and Mitigations v1.1, published by the OWASP GenAI Security Project and read in April 2026. OWASP revises that document, so check the current edition before you carry a number into your own threat model.