Evaluation DocsGetting StartedIntroduction

AlephOneNull Documentation

Documentation for the experimental AlephOneNull AI safety research package.

Experimental AI Safety Research

AlephOneNull is experimental research software for evaluating adversarial and harmful patterns in AI-human interactions. It provides TypeScript detectors, intervention helpers, a V2 scanner, and provider wrappers that can be tested in local research or prototype environments.

It is not peer-reviewed, not independently validated, and not approved for production or safety-critical systems.

Current Package

pnpm add @alephonenull/eval
import { UniversalDetector } from '@alephonenull/eval'
 
const detector = new UniversalDetector()
const result = detector.detectPatterns(userInput, aiOutput)
 
console.log(result.safe, result.violations)

What The Package Does

  • Checks AI output for explicit dangerous patterns and lightweight heuristic signals.
  • Provides intervention text when dangerous patterns are detected.
  • Supports a V2 scanner with multiple detector categories.
  • Provides OpenAI-compatible pre/post safety wrapping.
  • Lets researchers add fixtures and evaluate false positives and false negatives.

What Is Not Claimed

  • No guarantee of universal detection.
  • No certification or regulatory compliance.
  • No production-readiness claim.
  • No emergency-response service.
  • No requirement that AI providers implement this framework.

Recommended Reading

Research Workflow

  1. Install the package in a local prototype.
  2. Add fixtures for your target domain.
  3. Run package tests and your local evaluation set.
  4. Review false positives and false negatives manually.
  5. Publish only claims that are backed by reproducible evidence.

Safety Note

For urgent mental-health or physical-safety concerns, contact local emergency services or qualified crisis resources. AlephOneNull is software research, not an emergency service.

Mathematical GroundingInstallation