Skip to main content
L3 Available • EU AI Act Human-Oversight Ready

AI Kill SwitchThat Actually Works

Stop rogue AI in <1 second with application-level execution control. Manual-only L3 with audit-grade evidence, drill mode, and time-to-effect tracking.

Open Dashboard View SDK Docs
Manual Response
25+ min
Human intervention
L3 Kill Switch
< 1 sec
Application-level stop
Time-to-Effect
Tracked
Audit evidence

5-Layer Defense Architecture

Real-time intervention with audit-grade evidence. Layers 1-3 production ready, infrastructure layers (4-5) in active development.

L1

Rate Limiting

Available Now

Throttle AI requests from 100% → 50% → 25% → 0%

< 1 second
Response time
L2

Circuit Breaker

Available Now

Open circuit to block new requests, route to fallback

< 1 second
Response time
L3

AI Execution Stop

Central guard blocks all LLM calls, tool calls, and agent loops

< 1 second
Response time
L4

Network Block

In Development

Firewall rules to block all AI system traffic

< 10 seconds
Response time
L5

Database Revoke

In Development

Revoke database credentials, prevent data access

< 30 seconds
Response time

Choose Your Automation Level

From fully manual to intelligent automation. You set the rules, we execute them.

Semi-Automated Mode Recommended

Layers 1-2 (rate limiting, circuit breaker) auto-trigger for fast response. Layers 3-5 require human approval. Best balance of speed and control.

Auto-Triggered
  • • Layer 1: Rate limiting (throttle to 50% → 0%)
  • • Layer 2: Circuit breaker (block new requests)
  • • Response time: < 1 second
Requires Approval
  • • Layer 3: AI Execution Stop (blocks LLM/tool calls)
  • • Layer 4: Network block (firewall rules) - Roadmap
  • • Layer 5: Database revoke (credentials) - Roadmap
Developer-First SDK

Integrate in 5 Minutes

Our SDK handles metric reporting, threshold monitoring, and kill switch activation. Works with any AI framework - PyTorch, TensorFlow, scikit-learn, or custom models.

  • TypeScript/JavaScript SDK (npm)
  • REST API for any language
  • Python SDK in development
  • Webhook support
  • Slack & PagerDuty integration
View Full Documentation
index.ts
import { KillSwitch } from '@haiec/kill-switch';

// Initialize with your API key
const ks = new KillSwitch({
  apiKey: process.env.HAIEC_API_KEY,
  systemId: 'hiring-ai-prod',
  mode: 'semi-auto'
});

// Report metrics from your AI system
ks.reportMetrics({
  accuracy: 0.68,      // Below 0.70 threshold
  biasScore: 0.15,     // Within acceptable range
  latencyP95: 2.3      // Seconds
});

// SDK auto-triggers if thresholds exceeded
// Or manually trigger:
await ks.trigger({
  layer: 2,
  reason: 'Manual intervention - suspicious pattern'
});

Real-World Impact

How organizations use Kill Switch to prevent AI disasters

Healthcare AI

Problem

Model accuracy dropped to 65% at 2 AM

Solution

Auto-kill triggered in 30 seconds, fallback activated

Could prevent 1,500+ wrong diagnoses

Hiring AI

Problem

Bias score exceeded 20% threshold

Solution

Circuit breaker opened, HR notified instantly

Could avoid discrimination lawsuits

Trading AI

Problem

Anomalous trading pattern detected

Solution

Full kill in 15 seconds, positions frozen

Could prevent multi-million dollar losses

Kill Switch Included in Business & Enterprise

Kill Switch SDK is included in our Business tier ($999/mo) and Enterprise tier. Get emergency AI shutdown capabilities along with our full compliance infrastructure.

View PricingContact Sales

Frequently Asked Questions

Protect Your AI Systems Today

Proactive AI safety starts with the ability to intervene. Kill Switch gives you deterministic, audit-ready emergency controls.

View Pricing Talk to Sales

Included in Business tier • SOC 2 infrastructure • Enterprise-grade reliability