Thoth · AI Agent Governance · Aten Security

When your agent goes off script,the action doesn't execute.

Thoth enforces compliance policies on AI agent actions in real time, giving your security team fleet-wide visibility across every agent, model, and tool call.

Shadow mode free · No credit card · 7-day report included

Fleet dashboard: live view

47
Agents Active
12,483
Tool Calls / 24h
3
Blocked Today
billing-agent
2,341 calls
ENFORCE · HIPAA
data-export-agent
PHI bulk export blocked
BLOCKED
claude-desktop:nyah
1,104 calls
MCP · ENFORCE
hr-lookup-agent
891 calls
SHADOW

Every other tool sees before or after.
Thoth acts during.

The gap in enterprise AI security isn't a lack of tools. It's a lack of runtime.

Posture & Inventory

What exists

Configuration snapshots. Risk scores. Audit of what agents have access to.

Aim · Noma · Zenity · Astrix

Identity & Policy

What's permitted

Credentials issued. Tool access authorized. Stops at the policy gate.

Okta FGA · CyberArk · Aembit

Runtime Enforcement

What executes or doesn't

Behavioral enforcement at the moment of action. Before your infrastructure is touched.

Aten Security

Observability

What happened after

Traces and logs. Quality scores. No security layer, no blocking.

LangSmith · Langfuse · Arize

The window is closing.

AI agents are already running in regulated environments. Most have no runtime enforcement layer. Regulators are catching up fast.

57%

of enterprises have AI agents in production today

Gartner, 2025

82:1

machine vs. human identities. Zero with a runtime layer.

CrowdStrike, 2026

492

unauthenticated MCP servers found in production, Q1 2026

Wiz Research, 2026

Aug 2, 2026

Deadline for EU AI Act Article 12/13 enforcement begins

WORM evidence mandatory

Manage your entire AI agent fleet from one dashboard.

Real-time visibility into every tool call, every model invocation, every blocked action, across all your agents, at any scale.

Thoth Fleet Dashboard: prod
47
Agents
12,483
Tool Calls / 24h
3
Blocked
billing-agent
2,341 calls
ENFORCE · HIPAA
data-export-agent
PHI bulk export blocked
BLOCKED
claude-desktop:nyah
1,104 calls
MCP · ENFORCE
claude-desktop:jason
893 calls
MCP · ENFORCE
hr-lookup-agent
891 calls
SHADOW
  • Full fleet visibility

    Every agent in your organization, including custom pipelines, Claude Desktop, Cursor, and API calls, in one real-time view. No blind spots.

  • Block before it hits production

    Sub-100ms enforcement. When an agent tries to export PHI in bulk or call an out-of-scope tool, Thoth blocks it before the action executes.

  • Compliance-ready evidence

    Every blocked action generates a tamper-proof evidence bundle: immutable, hash-chained, ready for regulators the day they ask.

  • Progressive rollout

    Start in shadow mode. Run a 7-day baseline. Flip enforcement agent by agent at your pace. No big-bang cutover.

Your employees are using AI clients.
Thoth makes them enterprise‑safe.

The Thoth MCP proxy sits between Claude Desktop and your internal tools, giving security teams full visibility and enforcement over every tool call, without changing the user experience.

Before Thoth

Claude Desktop
MCP Server
EHR / CRM / DB

No visibility. No enforcement. No audit trail.

With Thoth MCP Proxy

Claude Desktop
Thoth Proxy
MCP Server
✓ Policy checked✓ Fleet logged✓ WORM receipt
Works with any MCP-compatible clientClaude Desktop · Cursor · Windsurf · Custom agents
  • Zero changes for end users

    Employees keep using Claude Desktop exactly as before. Thoth sits transparently in the request path, invisible to the user and visible to security.

  • Fleet-wide policy enforcement

    Apply HIPAA, SOC 2, or custom policies across every Claude Desktop instance in your organization simultaneously. One policy config, everywhere.

  • Deploys in under 30 minutes

    Point Claude Desktop at the Thoth MCP proxy endpoint. Configure your policy pack. Done. Your fleet is governed.

  • Minimum-necessary enforcement

    Thoth's HIPAA pack automatically applies minimum-necessary access rules at the tool level, blocking over-broad PHI access before it reaches your data layer.

Instrument once. Govern everything.

Three lines of code to instrument any agent. Pre-built policy packs for every major compliance framework. No custom rules required.

Developer SDK

Instrument your agent

Works with LangChain, CrewAI, Anthropic SDK, OpenAI, and any framework.

# instrument any agent in 3 lines

from thoth import instrument


instrument(

  agent,

  agent_id="billing-agent",

  policy_pack="hipaa-phi-agent",

  enforcement="progressive",

)


# Thoth handles enforcement, logging, WORM receipts

pip install aten-thothnpm i @atensec/thothgo get github.com/atensecurity/thoth-go

Compliance Policy Packs

Pick your framework

Runtime-enforced, not documentation. Activate a pack, Thoth enforces it.

HIPAA

HIPAA / HITECH

Step-up on PHI export · WORM chain

SOC 2

SOC 2 Type II

Cross-tenant block · Audit evidence

EU AI ACT

EU AI Act

Scope drift block · Art. 12 WORM

GDPR

GDPR

Cross-border block · Art. 30 records

NIST

NIST AI RMF

Human oversight gates · Baselines

OWASP

OWASP LLM Top 10

Injection defense · Excessive agency

+ FedRAMP · ISO 42001 · NIST 800-53 · HITECH

The CISO view.

We've seen an 800% increase in productivity agent usage. Post-approval monitoring is the critical gap nobody's solved.

CISO · Global Energy Company

When AI agents act on our behalf, identity is where governance either holds or collapses.

CISO · Major Professional Sports League

Post-approval governance of any access is a high-priority risk area for us.

CISO · Global Alternative Asset Manager

Advised by

GP · Ex-CISO

Caleb Sima

White Rabbit VC

Ex-CISO Robinhood · Databricks

CISO

Frederick Lee

Reddit

Ex-CISO Gusto · Twilio

CISO

Shaun Blackburn

True Anomaly

Ex-CISO Gemini · Airbnb · Netflix

SVP

Prem Iyer

Palo Alto Networks

Enterprise GTM Advisor

Your agents act autonomously.
So does Thoth.

Start in shadow mode. Zero risk. First behavioral report in 7 days. Then decide.