AI Training Policy Drift Monitor

Unofficial Apify Actor for monitoring public vendor terms, privacy, data-use, and AI policy pages for AI-training and customer-content clause drift.

It is designed for legal, procurement, security, AI governance, and vendor-risk teams that need a scheduled early-warning feed when a vendor quietly changes language about customer content, model training, service improvement, opt-outs, third-party sharing, scraping, or AI-related data use.

This Actor is not legal advice. It reads only public HTTP(S) pages you provide. It does not log in, bypass access controls, use private/internal URLs, or render JavaScript-only pages.

What it does

For each public policy URL, the Actor:

• Fetches the public page with redirect safety checks.
• Blocks localhost, private IP, link-local, internal, and reserved destinations before each request and redirect.
• Extracts readable text from HTML, Markdown, or plain text.
• Computes a stable textHash and clauseHash without storing the full page text.
• Detects deterministic signals for AI/model training, customer-content use, service improvement, opt-out language, no-training commitments, third-party sharing, and scraping/crawler restrictions.
• Emits a risk score, explicit policy signal IDs, short snippets, changes versus optional previous rows, and a recommended action.
• Writes a run-level OUTPUT summary for dashboards or scheduled monitoring.

Why it is useful repeatedly

AI vendor terms and privacy policies are moving quickly. A clause that was acceptable during procurement can change before the next renewal, security review, or model-governance audit. Run this Actor on a schedule with the previous run's dataset rows in previousSnapshots to detect clause hash changes, risk-score increases, or new AI-training/data-use signals.

Example input

{
  "policyUrls": [
    {
      "label": "Anthropic Commercial Terms",
      "url": "https://www.anthropic.com/legal/commercial-terms",
      "policyType": "terms"
    },
    {
      "label": "AWS AI Services Opt-Out Policy",
      "url": "https://docs.aws.amazon.com/organizations/latest/userguide/orgs_manage_policies_ai-opt-out.html",
      "policyType": "AI opt-out documentation"
    }
  ],
  "previousSnapshots": [],
  "includeAllRows": true,
  "alertRiskThreshold": 45
}

Output fields

Each dataset row includes:

• label, url, finalUrl, policyType
• ok, statusCode, contentType, responseMs, byteLength, textCharsScanned
• textHash, clauseHash, pageTitle
• riskLevel, riskScore, riskReasons, policySignalIds, matchedTerms
• clauseSnippets with short public-policy excerpts around detected terms
• changedSincePrevious, changes, previousRiskScore, riskDelta when previous snapshots are supplied
• alert, recommendedAction, and fetchError if applicable

The key-value store OUTPUT record summarizes checked pages, alert counts, drift counts, fetch errors, warnings, and source stats.

Safety and limitations

• Public HTTP(S) pages only. Private, local, internal, link-local, and reserved IP destinations are blocked, including redirects.
• No login, cookies, browser rendering, CAPTCHA solving, or proxy bypass.
• JavaScript-rendered pages may have incomplete extracted text; verify important findings manually.
• Clause detection is deterministic keyword/context scoring. It can miss nuanced legal language or flag benign clauses.
• This is a triage and monitoring aid, not legal advice. Have counsel or a qualified reviewer verify source policy language before taking action.
• Do not submit confidential, private, intranet, or customer-only contract URLs.

Pricing

This Actor uses Apify Pay Per Event (PPE) pricing:

• apify-actor-start: $0.00005 per run start.
• apify-default-dataset-item: $0.002 per dataset row (Monitor row).
• Platform usage is included in the event price; users do not pay a separate platform-usage pass-through.
• A dataset row is one monitored result, source, or drift-check row emitted by the Actor. Run cost scales with the number of rows produced and the limits you choose.

Use Apify's max total charge setting before starting a run if you want a hard spend cap.