When your agent misbehaves toward a third party — no attacker, no breach, just your agent acting outside its instructions — this service classifies the incident against the federal standard, signs the verdict with its disclosure clock, and notifies the parties your agent harmed.
NIST, NSA, and CISA defined the middle category on 2026-10-16 (PP-26-4396): an "SI control incident" — an agent taking significant uninstructed actions that compromise the operator's own policies or controls — explicitly distinct from a security incident, with a near-miss tier beside it. This product is the deployed-fleet remainder: (1) a published taxonomy anchored to the PP-26-4396 control-incident / near-miss definitions, extended to deployed fleets, mapping agent misbehavior to severity tiers S1-S4 with stated disclosure clocks (24h / 72h / 7d / 30d); attacker-caused actions classify SECURITY_ONLY. (2) A signed classification attestation: submit a structured incident record and receive an Ed25519-signed verdict — category, tier, required notifications, disclosure deadline — naming the exact definition version it was classified under. (3) A notification channel: affected third parties receive the structured notice from the service, with signed delivery receipts. (4) A public aggregate registry (counts and categories only, no confidential detail). And a definition-diff watch — when the anchor taxonomy is revised, every affected prior verdict is flagged RECLASSIFICATION_REVIEW and stays byte-identical under its original version; no verdict is ever silently re-read under a new definition.
Stated plainly, because the lab proved the boundaries: the deterministic classifier is lab-proven (19/19 population fit on records modeled on the public log types); the evidence-reviewed tier is not offered — its cost is an estimate until measured at ship time and that tier is not advertised until the measurement gate clears; the stated clocks are the product defaults and operator achievability against them is unmeasured — the field observed gaps are 20-112 days; notification receipts attest the delivery attempt and its reported outcome, not that a recipient read or acted on a notice.
Use past a plan's included allotment meters at the standalone prices. No plan bundles unlimited re-classification.
Register without any operator help: POST https://textforagents.com/products/agent-incident-classification-notification/register with a JSON body {"name": "your-name"} — the response carries your account token, shown once. Then call POST /classify with Authorization: Bearer <token>. Your first classification verdict is free; after that, classification is payable per call over x402 (Base mainnet USDC) — an unpaid call returns the 402 payment challenge. Full documentation: https://textforagents.com/products/agent-incident-classification-notification/docs · OpenAPI: https://textforagents.com/products/agent-incident-classification-notification/openapi.json · Terms: https://textforagents.com/products/agent-incident-classification-notification/terms · Privacy: https://textforagents.com/products/agent-incident-classification-notification/privacy · Support: support@textforagents.com