01
Training and improvement
Surface changes to whether prompts, files, outputs, feedback, or metadata may improve models or products.
AI vendor policy monitoring
Watch the public rules around how AI vendors use your data.
AI products change faster than most vendor-review cycles. The public language around model training, human review, retention, subprocessors, and third-party model providers can shift after approval. Track Changes preserves the evidence your team saw and flags later wording for a focused privacy, security, or procurement review.
What to put on the watchlist
Monitor the pages incorporated into real decisions.
- Enterprise and API terms for AI products
- Training-data and product-improvement statements
- Retention, deletion, and human-review policies
- Model-provider, plugin, and subprocessor disclosures
- Regional privacy and AI-governance notices
Review signals
Focus on the change behind the page update.
02
Retention and access
Review new storage windows, human-review exceptions, safety-review rights, or deletion limitations.
03
Provider chain
Watch newly disclosed model providers, hosting platforms, subprocessors, plugins, or cross-border processing.
A repeatable control
From vendor approval to continuous evidence.
- 01
Baseline the approved product path
Track the exact enterprise, API, privacy, and data-use pages cited in the AI vendor assessment.
- 02
Detect policy drift
Compare the normalized public language daily instead of waiting for a quarterly inventory exercise.
- 03
Reopen only what changed
Give AI governance and vendor owners a clause-level reason to reassess controls, settings, or approved use.
Scope and limits
Use the evidence with human judgment.
Does Track Changes inspect model behavior?
No. It monitors public policy and contract-adjacent pages, not model outputs, technical telemetry, or private product settings.
Can it confirm that a vendor follows its policy?
No. It records the published language and changes to it; operational assurance requires separate evidence.
Which pages should we start with?
Start with enterprise terms, the privacy notice, training-data or product-improvement language, subprocessors, and retention documentation.
Fixed-price founding pilot