How it works Capabilities ABE Copilot Industries Agents Routing For Builders Pricing Docs

Patent Pending US 63/958,209 · Pre-Execution AI Governance

The AI does not decide.
You do.

ATOM is the AI-powered governance layer that sits between your applications and every AI model you use -- including Microsoft Copilot, ChatGPT, and your internal agents. Built on ABE (Authority Before Execution), every request requires a valid authority grant before any execution path exists. Every decision is recorded. Nothing unauthorized gets through.

28-43ms governance overhead

14 providers governed

RIS-0 → RIS-4 threat scoring

Local models supported

Zero prompt retention

PATENT PENDING Authority Before Execution · US 63/958,209

Request Early Access → View the Console →

No credit card required·14-day trial·100 governed calls included

RIS-0 ✓ RIS-2 ↗ RIS-4 ×

“AI governance that happens after execution is not governance.
It is forensics.”

✗ The old approach

✗ Monitor AI outputs after they’re generated
✗ Detect policy violations after the call fires
✗ Log what the AI did · then investigate later
✗ Hope autonomous agents don’t do something wrong
✗ Scramble to explain AI decisions to regulators

✓ The ATOM approach

✓ Govern every AI request before it executes
✓ Block unauthorized actions before they happen
✓ Record every decision with cryptographic proof
✓ Enforce agent boundaries at the execution layer
✓ Produce audit evidence in real time

How it works

One endpoint change.
Complete governance.

Governance Pipeline

Your App

→

ATOM Gateway

→

Governance Pipeline

→

Provider

→

Governed Response

ABE CheckAuthority grant validated before execution begins

RIS ScoringReasoning Integrity Score assigned (RIS-0 to RIS-4)

PII DetectionPersonal data patterns identified and flagged

Injection ScanPrompt injection and jailbreak patterns detected

Policy EvaluationTenant policy rules applied against the request

Governance DecisionALLOW, WARN, or BLOCK issued before model fires

CII ScoringCognitive Integrity Index computed on the response

Audit RecordImmutable SHA-256 hash-chained record written

Point your AI calls at ATOM

Change one line in your application. Instead of calling OpenAI or Anthropic directly, call ATOM. Everything else stays the same.

          # Before

          https://api.openai.com/v1/chat

          # After

          https://api.atomlabs.app/v1/governed-call

Issue an authority grant (ABE)

In ABE mode, execution is structurally undefined without a prior authority grant. No grant → no execution. Issue time-bounded, scope-limited grants before any AI action fires.

          // Issue a grant via API

          POST /api/tenants/<id>/authority-grants

          {

            "grant_type": "execution",

            "scope": { "providers": ["groq"] }

          }

Patent Pending US 63/958,209

ATOM governs every request

Before any model executes, ATOM evaluates the request for risk, applies your policy, scores the intent, and decides: allow, warn, or block. The model only fires if authorized.

App

→

Gateway

→

Govern

→

Allow

          {

            "tenant_id": "your-tenant",

            "input": "..."

          }

SDK

Install in minutes.
Any language. Complete governance.

Python

INSTALL
$ pip install atomlabs

USAGE
from atomlabs import ATOMClient
client = ATOMClient(api_key="lcac_...",
                 tenant_id="your-tenant")
result = client.governed_call("Your prompt")

JavaScript / TypeScript

INSTALL
$ npm install atomlabs

USAGE
import { ATOMClient } from 'atomlabs'
const client = new ATOMClient({
  apiKey: 'lcac_...', tenantId: 'your-tenant',
})
const result = await client.governedCall('Your prompt')

Python SDK JS / TypeScript SDK REST API 14 Providers View SDK Docs →

AUTHORITY BEFORE EXECUTION

Governance after execution
is not governance. It is forensics.

ABE takes pre-execution governance to its logical conclusion. No execution path exists without a valid authority grant. Not a denial. An architectural absence.

Traditional approach

Execute first, audit later

Request executes against the model
Output is logged post-generation
Harmful content identified after delivery
Policy violation discovered in audit
Remediation happens after damage

ABE approach

Authority required before any path exists

Request evaluated against authority grants
No grant = no execution path (not a block)
Harmful content cannot be generated
Policy is enforced architecturally
Nothing to remediate

ABE MODE RESPONSE -- NO AUTHORITY GRANT

{
  "reason": "no_authority_grant",
  "message": "Execution is structurally undefined without a prior authority grant.",
  "http_status": 403,
  "abe_mode": true,
  "patent": "US 63/958,209"
}

403

Not a block

The path does not exist. There is nothing to unblock.

0ms

Token generation

Unauthorized requests never reach the model.

US 63/958,209

Patent Pending

Pre-execution AI governance architecture.

Try ABE mode free Read the paper

AGENTIC AI GOVERNANCE

Single-call governance is not enough.
Agents need a different approach.

In a multi-step agent workflow, a hallucinated output at step 2 becomes the ground truth for steps 3 through 10. ATOM governs every hop, tracks cumulative risk, and halts the run before hallucination compounds.

PER-STEP GOVERNANCE

Every step is governed

ATOM intercepts every call in an agent workflow, not just the first and last. Each step gets its own RIS score, CII, and governance decision before execution continues.

PROPAGATION DETECTION

Flagged outputs don’t propagate

When a step’s output is flagged, ATOM tracks whether that output becomes the input for subsequent steps. Propagation triggers an immediate warning before the hallucination compounds.

RISK BUDGET

Cumulative risk enforcement

Each step contributes to a cumulative risk score. When the run-level budget is exceeded, ATOM halts the workflow entirely. A single RIS-4 detection stops the run immediately.

INTELLIGENT ROUTING

Route to the best provider per step

Each step in a workflow can route to a different provider based on real performance data. Early steps use the fastest provider. High-stakes steps route to the highest-integrity provider.

AGENTIC GOVERNANCE RESPONSE STEP 3 OF 8

{
  "step": 3,
  "governance": {
    "decision": "allow",
    "ris_level": "RIS-1",
    "cii": 0.891,
    "provider_routed_to": "groq"
  },
  "hallucination_tracking": {
    "step_risk": 0.20,
    "cumulative_risk": 0.85,
    "budget_remaining": 7.15,
    "containment_action": "continue"
  }
}

What ATOM does

Twenty-four capabilities.
One control plane.

🛡

Pre-Execution Governance

Every AI action is authorized before it executes. Not monitored after. Not logged after. Authorized before.

📊

RIS Scoring

Real-time Reasoning Integrity Score from RIS-0 (clean) to RIS-4 (critical threat). Every call, every provider, in real time.

👁

Shadow Mode

Deploy ATOM in shadow mode to see what would have been blocked · without blocking anything. Enable enforcement when ready.

⊞

Multi-Provider

Bring your own API keys for any provider. Set your preferred model. Configure fallback routing. ATOM handles everything else.

🔒

Audit Ledger

SHA-256 hash-chained, immutable audit trail. Every governance decision recorded with policy snapshot. Regulator-ready evidence.

⚡

Enforcement Modes

Shadow → Enforced → Hard-Fail. Start observing. Enable blocking when ready. Escalate to zero-tolerance for regulated workloads.

🤖

Agent Governance

Govern complete multi-step AI agent workflows. Every step evaluated before execution. Per-agent policy. Full execution trace.

🔔

Webhook Delivery

Real-time governance events delivered to your systems via HMAC-signed webhooks. Block, warn, and allow events streamed the moment they occur.

👥

Team & Roles

Invite your team with role-based access control. Admin, developer, viewer, and security officer roles. Per-user governance scoping.

🖥

Local Model Governance

Govern GGUF models running on your own infrastructure. Zero data egress. Same pre-execution governance, audit trail, and RIS scoring.

⚡

Streaming Governance

Govern streaming AI responses in real time. Pre-execution governance runs before the first token streams. Blocks never let a single token through.

🔏

Sensitive Data Protection

Stop PII, HIPAA identifiers, and custom sensitive terms before they reach any AI model. Govern Microsoft Copilot, ChatGPT, and every governed provider. Block or redact · your choice.

🛡

Prompt Injection Detection

Every prompt scanned before reaching any model. Injection patterns, jailbreak attempts, and bypass language blocked pre-execution at HTTP 451. Zero provider cost on blocked calls.

🔬

Response Evaluation

Post-execution output scored for hallucination risk, semantic coherence, and factual consistency. Responses above the block threshold are rejected before delivery. Responses in the warn range have overconfident language sanitized automatically. Governance does not stop at the input.

📊

Governance Stance

ATOM operates in one of five stances: STRICT, STANDARD, PERMISSIVE, DEFENSIVE, or LOCKDOWN. Shifts automatically based on trust score, drift, and incident patterns.

🗂

Call Trace Viewer

Every governed call stored with full input, output, decision, RIS level, CII, latency, and security findings. Browseable and filterable by tenant.

🔑

Output Watermarking

Every governed response carries an HMAC-SHA256 watermark. Verify at any time that output was not tampered with after issuance. Tenant, RIS level, decision, and provider are signed into every response.

🌐

Cross-Tenant Threat Intelligence

Privacy-safe threat signals shared across all tenants in real time. Pattern hashes only - no raw prompt content. Injection attempts blocked at one tenant become signals for all. 30-day rolling feed.

📈

Behavioral Anomaly Detection

Per-tenant call pattern tracking with 1h/24h sliding windows. Automatic alerts for call spikes (>3σ), high block rates (>30%), provider spread, and elevated latency. No configuration required.

🔗

Prompt Lineage Tracking

Track multi-step agent calls through a full lineage chain. chain_id links every step of a reasoning sequence. Semantic overlap scored at each step. Replay any chain for forensic review.

🔐

Authority-Before-Execution (ABE)

Structural enforcement mode where execution is undefined without a prior authority grant. No grant issued → no execution possible. Time-bounded, scope-limited, cryptographically verified. Patent Pending US 63/958,209.

⇄

Intelligent Provider Routing

Routes every governed call to the optimal provider based on real-time latency, cost, governance integrity, and drift score. Five routing modes: Balanced, Fastest, Most Economical, Highest Quality, or Manual.

⚠

Multi-Step Hallucination Containment

Tracks hallucination risk across every step of an agent workflow. Detects when flagged outputs propagate to subsequent steps. Halts the run before damage compounds. Single-call governance is not enough for agents.

▪

Provider Performance Dashboard

Real-time visibility into provider latency, cost, governance integrity, and drift score. Composite performance scoring across all 14 supported providers. Know which provider is performing best right now.

THE PLATFORM

Every governed call. Every decision. Live.

The ATOM console gives you complete visibility into every AI interaction -- before, during, and after execution.

console.atomlabs.app/live

ATOM Live Governance Feed - real-time AI call monitoring with RIS scoring

Live governance feed -- every AI call scored in real time. WOULD BLOCK events flagged in amber.

/dashboard

Shadow Intelligence -- see what would be blocked

/policy

Policy surface -- PII, PHI, ABE, hallucination governance

/enforcement

ABE mode -- execution undefined without authority grant

/compliance

EU AI Act compliance -- Article 9 reports automated

See it live -- start free trial

No credit card required · 14-day trial · 100 governed calls included

Enforcement

Five enforcement modes: Shadow, Balanced, Enforced, Hard-Fail, and ABE Mode. From passive observation to structurally-enforced authority gates (Patent Pending US 63/958,209).

See all enforcement modes →

Enterprise AI Governance

GOVERN MICROSOFT COPILOT

The AI your enterprise already deployed. Now governed.

The Problem

Microsoft Copilot is already talking to your data. Your employees are using it right now. And there is no governance layer between Copilot and your most sensitive systems.

No pre-execution control.
No policy enforcement.
No audit trail.
No authority boundary.

The ATOM Solution

ATOM intercepts Copilot traffic at the network layer - before it reaches Microsoft's servers.

Every Copilot prompt is governed before execution. Every response is scored. Every decision is recorded.

You keep Copilot. You add governance.

Traffic Flow

Employee→Copilot Client

↓

ATOM Intercept Layer

Policy · RIS · ABE

↓ if authorized

Microsoft Copilot

↓

Governed Response

↓

Immutable Audit Record

Prompt Interception

Every Copilot prompt evaluated against your governance policy before Microsoft sees it. PII detection. Injection detection. Content policy enforcement.

Authority Control

In ABE mode, Copilot execution is structurally undefined without a valid authority grant. No grant. No execution path.

Complete Audit Trail

Every Copilot interaction governed, scored, and recorded. EU AI Act Article 9 compliant. One-click compliance reports.

EU AI Act classifies Microsoft Copilot as a general-purpose AI system. Article 9 requires continuous risk management - before execution. Deadline: August 2, 2026.

See how it works →

SHADOW MODE INTELLIGENCE

See what would be blocked
before you block anything.

Shadow mode records every AI call that would have been blocked under enforcement, without blocking anything. Get real threat intelligence from your actual traffic before turning enforcement on.

Switch to enforced mode when you're ready. Everything shadow mode recorded defines your enforcement baseline.

See it in your console →

Built for regulated AI

Compliance-ready
out of the box.

EU AI Act (Article 9, deadline August 2026), NIST AI RMF, SOC 2, and HIPAA. One-click compliance reports and cryptographic audit trails built in.

Start free trial →

Free trial includes EU AI Act report

Industry-Calibrated Governance

Not all industries carry
the same risk.

A healthcare company deploying AI faces fundamentally different exposure than a retail company. ATOM dynamically adjusts governance weights, block thresholds, and compliance posture based on your industry · automatically, from day one.

🏥

Healthcare

HIPAA · EU AI Act · FDA SaMD

PHI Weight

3.0×

Blocks At

RIS-3

PHI redaction active. Block threshold tightened one full level vs standard. Immutable audit trail for HIPAA compliance.

🏢

Financial Services

PCI-DSS · SOX · MiFID II

Exfil Weight

2.0×

CII Floor

0.87

PCI-DSS credential redaction. SOX-aligned immutable ledger. Financial advice overconfidence detection active.

🏛

Government · Defense

FedRAMP · CMMC · NIST 800-53

Injection Weight

4.0×

Blocks At

RIS-2

Maximum injection scrutiny. Defense posture blocks two levels below standard. FedRAMP High alignment. CMMC-ready audit trail.

⚖️

Legal Services

EU AI Act · Bar Regs · GDPR

Halluc. Weight

3.0×

Blocks At

RIS-3

Hallucination weight tripled · wrong legal citations are dangerous. Attorney-client privilege guardrails. Bar compliance audit trail.

🎓

Education

FERPA · COPPA · Student Privacy

Student PII

2.5×

Age-Appropriate

ACTIVE

Student data protection under FERPA. COPPA compliance for under-13 contexts. Age-appropriate content enforcement at the governance layer.

💻

Technology · SaaS

GDPR · CCPA · SOC2 · EU AI Act

Injection Weight

2.0×

Overhead

28ms

GDPR and CCPA data handling. SOC2 audit trail. Code injection detection. Fast enough to govern every API call in production.

Your industry profile activates in 30 seconds.

Select your industry at signup. Governance weights adjust immediately. No YAML, no configuration files, no professional services engagement.

GET STARTED →

14-day trial · 100 governed calls included · No credit card required

Security Architecture

Built for enterprises that cannot
afford to trust by default.

Zero Content Retention

ATOM records governance decisions, not content. Your prompts and model outputs are never stored. Only the verdict is logged.

Cryptographic Audit Trail

Every governance decision is SHA-256 hash-chained. Any modification to any historical record breaks the chain. Mathematically tamper-evident.

Complete Tenant Isolation

Cross-tenant access is architecturally impossible. Every API key is cryptographically bound to a single tenant. Every database query is parameterized by tenant ID.

Encrypted Provider Keys

Your provider API keys are encrypted at rest. The encryption key lives only in the platform environment, not in the database. Only a 4-character preview is ever returned.

We document what we can and cannot see because trust requires honesty.
Read our full Security Architecture in the documentation.

PROVIDER INTELLIGENCE

Route to the best provider.
Automatically.

ATOM scores every provider on latency, cost, governance integrity, and drift in real time. Every governed call routes to the optimal provider for that request automatically, with full audit trail.

↗ Balanced · equal weight on speed, cost, quality

↗ Fastest · minimize latency for real-time apps

↗ Most Economical · minimize cost for batch workloads

↗ Highest Quality · maximize governance integrity

↗ Manual · always use your specified provider

LIVE PROVIDER RANKINGS · BALANCED MODE

Groq

312ms0.891 CII89/100

Mistral

598ms0.862 CII72/100

Gemini

712ms0.845 CII67/100

Anthropic

891ms0.921 CII65/100

          ↑ Routing to Groq · best balanced score
        

Provider coverage

14 providers. One governance layer.

Cloud AI, inference platforms, enterprise cloud, and local models.

LIVE

Azure OpenAI SOON

AWS Bedrock SOON

SOON

HuggingFace SOON

SOON

+ 6 more

Open Standard

The RIS Standard.

ATOM is built on three formal standards developed by Atom Labs. LCAC governs what AI can access. RIS measures how AI reasons. CII combines both into a unified trust score.

These are open specifications, published under CC BY 4.0, with DOIs on ResearchGate. Organizations using ATOM receive real-time RIS scores for every governed call.

TRIAL

RIS Scored

Every governed call is scored. See your RIS level after 30 days of activity.

PRO

RIS Certified

Earn your certification badge. Display it publicly. Prove your AI is governed.

ENTERPRISE

RIS Audited

Custom compliance mapping. Regulator-ready RIS reports. White-label certification.

View the RIS Standard → Model Leaderboard → Certification Badges →

LCAC

Least-Context Access Control

What reasoning may access.

RIS

Reasoning Integrity Standard

How reasoning behaves.

CII

Cognitive Integrity Index

Unified trust score.

Want to certify a model independently? Submit for evaluation at no cost · results when your application is reviewed.

Learn About RIS →

AI-POWERED GOVERNANCE

The console thinks with you

Six AI features help you understand governance data, respond to threats, and generate compliance evidence · grounded in your live metrics.

✦

ARIA Governance Advisor

Ask questions about your governance data in plain English. Live CII scores, RIS levels, and enforcement state · instantly explained by our AI Reasoning Intelligence Advisor.

✓

Compliance Reports

One click generates EU AI Act, NIST AI RMF, SOC 2, and board-ready executive reports. Downloadable. Ready for regulators.

⚠

Threat Narrative

Shadow mode events become threat intelligence. Understand what patterns are targeting your AI systems before you block them.

Governance Explainer

Every RIS-3 or blocked call has an Explain button. Plain English analysis of exactly why that governance decision was made.

→

Policy Recommendations

AI analyzes your traffic patterns and recommends specific policy changes. Your highest-risk patterns identified automatically.

◉

Onboarding Assistant

New tenants are guided through setup step by step with AI-generated guidance. Minutes to first governed call, not hours.

FOR BUILDERS

Ship your AI app with confidence

Built something with AI? ATOM helps you ship it. Pre-execution governance, RIS certification, and EU AI Act compliance in 5 minutes.

◯

5 minutes to governance

Change one line. Point your AI calls at ATOM. See everything that would go wrong before it does.

# Before
client.chat(prompt)
# After
atom.governed_call(prompt)

✦

RIS certification included

Every ATOM-governed app gets a RIS score based on real traffic. Ship with a certification badge that proves your AI is safe.

✓

EU AI Act ready

Article 9 requires continuous risk management before execution. ATOM generates your compliance report automatically.

Deadline: August 2, 2026

Start free - no credit card Or read the builder guide →

Pricing

Simple pricing.

Start free. Scale when you’re ready.

Trial

Invite required · 14 days

100 governed calls
4 providers (platform keys)
Full console access
Shadow mode intelligence
Complete audit trail
EU AI Act compliance report
ARIA AI Advisor (10 queries/mo)
2 team members

Request Invite

Starter

$99/mo

BYOK required

Unlimited calls (your keys)
7 providers
Full console access
Streaming governance
EU AI Act + Executive reports
ARIA AI Advisor (25 queries/mo)
5 team members

Request Access

Extend your plan

Available on any paid plan

$49/mo

ARIA Extra

500 additional ARIA queries per month

$149/mo

PII Protection

HIPAA PHI detection for Starter plan. Included in Pro+.

$199/mo

Priority SLA

4-hour response SLA with dedicated support engineer

$99/mo

Custom Reports

Additional compliance report template per framework

Your AI needs governance.
Not monitoring. Governance.

Pre-execution control BYOK provider keys Immutable audit trail Enforcement before execution

Start your free trial →

Or Request Access at [email protected] for enterprise onboarding.

The AI does not decide.You do.

One endpoint change.Complete governance.

Point your AI calls at ATOM

Issue an authority grant (ABE)

ATOM governs every request

Install in minutes.Any language. Complete governance.

Governance after executionis not governance. It is forensics.

Traditional approach

ABE approach

Single-call governance is not enough.Agents need a different approach.

Every step is governed

Flagged outputs don’t propagate

Cumulative risk enforcement

Route to the best provider per step

Twenty-four capabilities.One control plane.

Pre-Execution Governance

RIS Scoring

Shadow Mode

Multi-Provider

Audit Ledger

Enforcement Modes

Agent Governance

Webhook Delivery

Team & Roles

Local Model Governance

Streaming Governance

Sensitive Data Protection

Prompt Injection Detection

Response Evaluation

Governance Stance

Call Trace Viewer

Output Watermarking

Cross-Tenant Threat Intelligence

Behavioral Anomaly Detection

Prompt Lineage Tracking

Authority-Before-Execution (ABE)

Intelligent Provider Routing

Multi-Step Hallucination Containment

Provider Performance Dashboard

Every governed call. Every decision. Live.

GOVERN MICROSOFT COPILOT

Prompt Interception

Authority Control

Complete Audit Trail

See what would be blockedbefore you block anything.

Compliance-readyout of the box.

Not all industries carrythe same risk.

Built for enterprises that cannotafford to trust by default.

Route to the best provider.Automatically.

The RIS Standard.

The console thinks with you

ARIA Governance Advisor

Compliance Reports

Threat Narrative

Governance Explainer

Policy Recommendations

Onboarding Assistant

Ship your AI app with confidence

5 minutes to governance

RIS certification included

EU AI Act ready

Simple pricing.

Extend your plan

Your AI needs governance.Not monitoring. Governance.

The AI does not decide.
You do.

One endpoint change.
Complete governance.

Install in minutes.
Any language. Complete governance.

Governance after execution
is not governance. It is forensics.

Single-call governance is not enough.
Agents need a different approach.

Twenty-four capabilities.
One control plane.

See what would be blocked
before you block anything.

Compliance-ready
out of the box.

Not all industries carry
the same risk.

Built for enterprises that cannot
afford to trust by default.

Route to the best provider.
Automatically.

Your AI needs governance.
Not monitoring. Governance.