Verbatim — AI Vendor Evidence Packets (Apify Actor)

The terms you signed off on are not the terms anymore.

Verbatim watches what AI vendors publish about your data, and quotes it back exactly. For each of 46 AI vendors (product + plan), this Actor returns exact quoted passages — each with its source URL and capture date — on six dimensions:

1. Training use of customer data
2. Retention
3. Deletion
4. Zero-data-retention
5. Processing regions
6. Subprocessors

Where a vendor's published documents say nothing about a dimension, the packet says so explicitly (NOT_SPECIFIED). Verbatim never paraphrases, infers, or fills gaps. Packets are evidence, not legal advice, not certification, and not a guarantee of vendor behaviour.

This Actor is a thin client over the public Verbatim REST API at https://api.trustcyber.ca (documentation: https://api.trustcyber.ca/docs).

Modes and pricing

Mode	Input	Result	Price
list_vendors	none	All 46 vendors (slug, vendor, product, plan), one dataset item each	Free
get_packet	vendorSlug, format	The vendor's latest evidence packet	US$0.10 per run
compare_packet	vendorSlug	Before/after quote pairs between the two most recent captures	US$0.05 per run
get_all_packets	none	Every vendor's latest evidence packet, one dataset item each (46 items)	US$2.99 per run

Pricing is pay-per-event, set by the pricing research enacted 2026-10-23 (replacing the original US$5 / US$12 guess). The events (packet-delivered at US$0.10, comparison-delivered at US$0.05, bundle-delivered at US$2.99) and their prices are declared in .actor/actor.json under pricingInfos (pricingModel: PAY_PER_EVENT), and the Actor calls Actor.charge(...) once per successful paid run. Platform usage is included in the event prices, the actor-start event stays at its platform default, and no Store discount tiers are enabled at launch, so the headline prices above are the real prices. list_vendors never charges. Failed or unknown-vendor runs are never charged. A comparison for a vendor with only one capture returns compared: false with an explanation instead of a diff, and is NOT charged: the comparison-delivered event fires only when a comparison actually ran. The prices take effect only when the Actor's owner confirms the pricing in the Apify Console (Monetization tab) on their own account; until then the values in actor.json are intent, not live billing.

Input

{
  "mode": "get_packet",
  "vendorSlug": "anthropic",
  "format": "json"
}

Field	Type	Default	Notes
mode	string (enum)	list_vendors	list_vendors | get_packet | compare_packet | get_all_packets
vendorSlug	string	—	Required for get_packet / compare_packet. A slug from list_vendors, e.g. anthropic, openai, microsoft-azure-openai.
format	string (enum)	json	get_packet only: json (structured packet) or markdown (rendered report)
apiBaseUrl	string	https://api.trustcyber.ca	Advanced: point at your own Verbatim instance

Output

Results go to the run's default dataset, and the primary payload is also stored under the OUTPUT key of the default key-value store.

• list_vendors: one dataset item per vendor: {"mode": "list_vendors", "slug": "anthropic", "vendor": "Anthropic", "product": "Claude (Commercial / API)", "plan": "Commercial terms (Team, Enterprise & API)"}
• get_packet (json): one dataset item, the full packet: {"vendor", "product", "plan", "slug", "headline", "captured_at", "disclaimer", "sources": [...], "findings": {"training_use": {"status": "FOUND", "passages": [{"quote", "source_url", "source_key", "captured_at"}]}, ...}}
• get_packet (markdown): one item {"mode": "get_packet", "slug", "format": "markdown", "markdown": "# Verbatim Evidence Packet: ..."}.
• compare_packet: one item with compared, from, to, change_count, and changes (before/after quote pairs); or compared: false plus an explanation when a second capture does not exist yet.
• get_all_packets: 46 dataset items, one full packet per vendor (same shape as get_packet JSON); the OUTPUT key holds {"mode": "get_all_packets", "count": 46, "packets": [...]}.

Examples

List the covered vendors (free):

{ "mode": "list_vendors" }

Get Anthropic's current packet as JSON (US$0.10):

{ "mode": "get_packet", "vendorSlug": "anthropic", "format": "json" }

Get OpenAI's packet as a Markdown report (US$0.10):

{ "mode": "get_packet", "vendorSlug": "openai", "format": "markdown" }

See what Microsoft Azure OpenAI changed between its last two captures (US$0.05):

{ "mode": "compare_packet", "vendorSlug": "microsoft-azure-openai" }

Get every vendor's packet in one run (US$2.99 bundle):

{ "mode": "get_all_packets" }

Sample packet (full, real)

One complete packet, exactly as get_packet returns it in Markdown form (Anthropic, captured 2026-10-20; the live version is at https://api.trustcyber.ca/packet/anthropic.md):

---

Verbatim Evidence Packet: Anthropic

The terms you signed off on are not the terms anymore.

• Product: Claude (Commercial / API)
• Plan scope: Commercial terms (Team, Enterprise & API)
• Captured: 2026-10-20

Verbatim packets report what vendors publish, quoted and dated. They are evidence, not legal advice, not certification, and not a guarantee of vendor behaviour.

Findings

Training use of customer data

See our Non-User Privacy Policy for more details on the data used to train our models, and the different stages of development.

Source: https://www.anthropic.com/legal/privacy (captured 2026-10-20)

Anthropic obtains personal data from third party sources in order to train our models.

Source: https://www.anthropic.com/legal/privacy (captured 2026-10-20)

Anthropic is a global company, and when you access our Services, your personal data is transferred to our servers in the US, or to other countries outside the European Economic Area (“EEA”) and the UK for the purposes outlined in Section 2 of this Privacy Policy, for example to provide you with the Service, or to train our models, or to conduct research.

Source: https://www.anthropic.com/legal/privacy (captured 2026-10-20)

Retention

Anthropic retains your personal data for as long as reasonably necessary for the purposes and criteria outlined in this Privacy Policy and explained further in our privacy center.

Source: https://www.anthropic.com/legal/privacy (captured 2026-10-20)

Deletion

You also are able to delete individual conversations, which will be removed immediately from your conversation history and automatically deleted from our back-end within 30 days.

Source: https://www.anthropic.com/legal/privacy (captured 2026-10-20)

H.1.b. delete all copies of Customer Data (including Customer Personal Data) processed by Anthropic or any Subprocessors, except to the extent (i) Applicable Data Protection Laws or other applicable legal or regulatory requirements requires storage of the Customer Data, (ii) retention of the Customer Data by Anthropic is necessary to resolve a dispute between the parties, or (iii) retention of the Customer Data is necessary to combat harmful use of the Services.

Source: https://www.anthropic.com/legal/data-processing-addendum (captured 2026-10-20)

Zero-data-retention

NOT SPECIFIED in the sources captured.

Processing regions

NOT SPECIFIED in the sources captured.

Subprocessors

By providing us with personal data, you acknowledge and agree that your personal data may be transferred or disclosed to other jurisdictions for processing and storage outside of Canada, including to the United States and the countries listed on our Subprocessor List, where laws regarding the protection of personal data may be less stringent than the laws in your jurisdiction.

Source: https://www.anthropic.com/legal/privacy (captured 2026-10-20)

"Subprocessor" means an entity engaged by Anthropic to process Customer Personal Data.

Source: https://www.anthropic.com/legal/data-processing-addendum (captured 2026-10-20)

Customer grants Anthropic general authorization to engage the Subprocessors listed in Schedule 4, and any additional Subprocessors in accordance with Section C.3. below.

Source: https://www.anthropic.com/legal/data-processing-addendum (captured 2026-10-20)

Sources captured

• Privacy policy: https://www.anthropic.com/legal/privacy [ok]
• Commercial terms of service: https://www.anthropic.com/legal/commercial-terms [ok]
• Data processing addendum: https://www.anthropic.com/legal/data-processing-addendum [ok]
• Subprocessor list: https://www.anthropic.com/subprocessors [empty - Only 0 chars of text extracted (JS-gated page?)]

---

Coverage, honestly

• 46 vendors, one product + plan each. Every quoted passage in the current packets was verified verbatim against the vendor's own published page: 268 of 268 quotes verified.
• Three vendors score 0 of 6 dimensions (Oracle AI, Sourcegraph Cody, Windsurf). That is not missing data: their published documents genuinely do not address these six dimensions, and their packets say so, dimension by dimension.
• Capture dates are printed in every packet. A packet is a dated snapshot of what the vendor published on that date, not a live claim about the vendor's current behaviour.

Local development

The API client (src/verbatim_client.py) uses only the Python standard library, so the core logic runs without the Apify SDK:

# Smoke test against the live API; writes evidence to test-output/
python3 smoke_test.py --compare anthropic

# Full Actor locally (requires the Apify CLI and an initialized storage dir)
pip install -r requirements.txt
apify run

test-output/ in this repo holds real outputs from a smoke-test run: list_vendors.json (46 vendors), get_packet-anthropic.json (the full Anthropic packet), compare_packet-anthropic.json, and summary.json.

Publishing (owner checklist — not done by this repo)

1. Create/confirm an Apify account and, in Apify Console → Settings → Integrations, create an API token. Enter the token only through the secure credential flow (apify login or CI secret); never commit it.
2. From this directory: apify push (builds the Dockerfile on Apify).
3. In the Actor's Console page: set the title/description (already in .actor/actor.json), upload a logo if desired, and set the Actor to Public if it should appear in Apify Store.
4. Monetization tab: confirm pay-per-event pricing — event packet-delivered at US$0.10, comparison-delivered at US$0.05, and bundle-delivered at US$2.99, list_vendors free — matching pricingInfos in .actor/actor.json. Confirm platform usage is set to included in the event prices, the actor-start event is at its platform default, and Store discount tiers are off. While in the Console, also confirm the payout method, payout currency, and whether a minimum payout threshold applies.
5. Run one paid-mode test on the platform and confirm the charge event appears in the run's billing before announcing the listing.

Layout

.actor/actor.json          Actor definition + pay-per-event pricing placeholders
.actor/input_schema.json   Console/API input form
.actor/output_schema.json  Dataset + OUTPUT declarations
.actor/dataset_schema.json Dataset fields and overview view
src/__main__.py            Entry: python3 -m src
src/main.py                Apify lifecycle: input -> API -> dataset/OUTPUT -> charge
src/verbatim_client.py     Verbatim REST client (stdlib only) + input validation
Dockerfile                 apify/actor-python:3.12
requirements.txt           apify SDK (platform runs only; the client needs nothing)
smoke_test.py              SDK-free local test against the live API
test-output/               Captured smoke-test evidence