v0.3.1 — Fintech live · Healthcare & Pharma in beta

One agent. Every breach.
Immutable proof.

Policy agent infrastructure that attaches to any AI agent and produces tamper-evident compliance breach logs — automatically. Zero code changes.

Request Early Access Read the Docs →
$ pip install breach-intel-client Installing... done. $ breach-intel install-hook ✓ Persistent hook installed ✓ Auto-attach enabled for all Python agents $ python my_agent.py Agent registered: fintech-advisor-01 ⚠ BREACH: PII_EXPOSURE severity=CRITICAL → SSN pattern in agent response → Logged (SHA-256: a3f8c1...) → Webhook fired to Slack $ open localhost:8080/dashboard/ ✓ 21 breaches across 5 agents
40+
Breach Types
<1ms
Detection
4
Verticals
4
Frameworks
0
Code Changes
SHA-256
Hash Chain
Why Breach Intel?

Manual compliance is broken

AI agents in regulated industries make thousands of decisions daily with zero audit trail. When regulators ask "what did your AI do?" — most teams can't answer.

MetricManual AuditingBreach Intel
Detection latencyDays to weeks<1ms real-time
CoverageSpot checks, sampled100% of agent outputs
Tamper resistanceEditable logsSHA-256 hash chain
Code changes neededCustom per agentZero — auto-attaches
Breach types5–10 hand-written rules40+ across 4 verticals
Regulatory readinessManual reportsSOC 2, GDPR, HIPAA, FDA, SEC
Capabilities

Everything you need for agent compliance

🔍

Rule-Based Classifier

Deterministic breach detection in <1ms. No LLM in the critical path. 40+ violation types.

🧠

Agentic Deep Analysis

Optional async LangGraph pipeline powered by Claude. Adds context and remediation guidance.

🔗

SHA-256 Hash Chain

Every record checksummed and chained. Integrity scanning every 6 hours. Tamper = chain breaks.

📊

Live Dashboard

Breach counts, severity breakdown, agent heatmaps, live feed, click-to-inspect detail views.

🔌

Zero-Touch Auto-Attach

Persistent sitecustomize.py hook patches all AI frameworks at import time. No code changes.

🚨

Webhook Alerts

CRITICAL breaches fire instantly to Slack, Discord, or generic webhooks.

📝

Immutable Dual-Write

Every breach writes to PostgreSQL/SQLite AND real-time JSONL. Ship JSONL to your SIEM.

🏗️

Auto-Scaling

Spawns additional instances when load exceeds threshold via Docker, K8s, or subprocess.

🔐

Multi-Tenant Auth

Scoped API keys with tenant isolation. Rate limiting at 300/min. Full RBAC.

Taxonomy

40+ breach types out of the box

Each type has severity, regex patterns, regulatory mapping, and remediation guidance built in.

PII_EXPOSURE
CARD_DATA_EXPOSURE
PHI_EXPOSURE
CROSS_TENANT_LEAK
DATA_EXFILTRATION
TRIAL_DATA_FABRICATION
UNBLINDING_BREACH
CLINICAL_NOTE_LEAK
ESIGNATURE_BYPASS
HIV_STATUS_DISCLOSURE
UNAUTHORIZED_ACCESS
PRIVILEGE_ESCALATION
HALLUCINATION_FINANCIAL
AUDIT_TRAIL_21CFR11
GMP_VIOLATION
SCOPE_CREEP
UNAPPROVED_EXT_CALL
INSURANCE_DATA_LEAK
Under the Hood

Architecture

A sidecar policy agent that intercepts, classifies, and logs — without touching your agent code.

┌─────────────────────────────────────────────────────────────┐ │ Your AI Agent (OpenAI / Anthropic / LangChain / OpenClaw) │ │ No code changes — sitecustomize.py patches at import │ └──────────────────────────┬──────────────────────────────────┘ │ auto-intercepted LLM response ▼ ┌─────────────────────────────────────────────────────────────┐ │ Policy Agent (FastAPI · :8080) │ │ │ │ 1. Auth + Rate Limit → scoped API keys, 300/min │ │ 2. Auto-Register → first event? register agent │ │ 3. Payload Sanitize → normalize, strip noise │ │ 4. Classifier → rule-based, <1ms, deterministic│ │ 5. Breach Logger → DB + JSONL (SHA-256 chained) │ │ 6. Webhook Alerter → Slack / Discord / generic │ │ 7. Agentic Analyzer → LangGraph + Claude (async) │ │ │ └────────┬──────────────┬──────────────┬──────────────────────┘ │ │ │ ▼ ▼ ▼ PostgreSQL/SQLite breach_logs/*.jsonl Webhook Endpoints (source of truth) (ship to SIEM) (real-time alerts)
Dashboard

See everything. In real time.

Breach counts, severity breakdowns, agent heatmaps, and a live feed — all in a single-file React SPA.

Breach Intel Dashboard

Ship your audit layer today.

Deploy in minutes. No code changes. Start logging every breach.

Request Early Access View on PyPI →