---
title: "Find exposure from breaches that look AI-agent-sourced"
method: POST
path: "/v1/metered/prompt-injection-breach"
tags: ["Agent and MCP security"]
---

# Find exposure from breaches that look AI-agent-sourced

`POST /v1/metered/prompt-injection-breach`

## Request

- Base URL: `https://api.relayshield.net`
- URL: `https://api.relayshield.net/v1/metered/prompt-injection-breach`
- Auth: one of:
  - API key in header `X-RS-API-KEY`
  - API key in header `X-API-Key`

Returns stolen session and credential records for an address where the source dump's own announcement text suggests an AI agent, rather than a traditional phishing or malware campaign, was involved in obtaining it.

**This is a heuristic, not an attribution.** It is a keyword classifier over dump-announcement text, with no labelled ground truth. A match is a reason to look closer, and no match does not rule out an AI-agent-involved breach. For unfiltered stolen-session exposure on the same address, use `/v1/metered/session-risk`.

**Billing.** $0.35 per successful call, metered to your card. Failed calls are never billed. Threat Intelligence subscribers and AWS Marketplace bundle customers call this at no extra charge. Unlike the other metered endpoints this one does not check a prepaid credit balance before running, so it never returns `402`.

**Note.** Reads the same corpus as `/v1/metered/session-risk`, filtered to records the ingestion pipeline flagged as suspected-agentic-source. Also billed as an AWS Marketplace Bundle D dimension for bundle customers.

## Request body

- object
  - `email` string, email, required: Email address to check.

## Response `200`

Success. `ok` is true and `data` holds the result.

- object
  - `ok` true, required
  - `data` object, required
    - `email` string: The normalised address that was checked.
    - `found` boolean: True when at least one record carries the suspected-agentic-source marker.
    - `session_count` integer: Number of matching records.
    - `sessions` object[]: Matching records. Unordered.
      - `domain` string: Service the stolen session or credential belongs to.
      - `severity` 'CRITICAL' | 'HIGH' | 'MEDIUM' | 'LOW': Severity derived from the service category.
      - `service_category` string: Category of the affected service, e.g. cloud_console, identity_provider, productivity.
      - `channel_source` string: Criminal channel or archive the record was collected from.
      - `ingested_at` string: When RelayShield ingested the record.
    - `note` string: Present only when `found` is false. States that a non-match does not rule out an AI-agent-involved breach.
    - `action_required` string: Present only when `found` is true. Remediation guidance.

## Other responses

- `400`: `email` is missing or contains no '@'.
- `401`: Missing or invalid API key.
- `405`: Wrong HTTP method. Every endpoint is POST.
- `429`: A shared demo key's daily quota is exhausted. Does not apply to your own key.
- `500`: Stolen-session corpus query failed.

---

[API](https://skmtc.dev/relayshield/apis/relayshield-api.md) · [All operations](https://skmtc.dev/relayshield/apis/relayshield-api/llms.txt) · [OpenAPI document](https://skmtc.dev/relayshield/apis/relayshield-api/revisions/4d4721b936dd?raw)