← Ataski · Trust & security · DPA · Subprocessors

This is the canonical English version. Native translations for DE / ES / FR / PT / RU are pending counsel sign-off.

Privacy Policy

Last updated: 2026-06-19 Effective: 2026-06-19 Status: Draft for legal review. This is the platform-wide Privacy Policy for Ataski (ataski.com). It supersedes the prior role-scoped scaffold and now describes how Ataski handles personal data across the entire service. It supplements, and cross-references but does not replace, the Terms of Service (docs/legal/terms-of-service.md, public: https://ataski.com/legal/terms), the Data Processing Addendum (docs/legal/dpa-template.md, public: https://ataski.com/legal/dpa), the Sub-Processors disclosure (docs/legal/subprocessors.md, public: https://ataski.com/legal/subprocessors), the Data Retention Policy (docs/legal/data-retention-policy.md, public: https://ataski.com/legal/data-retention), and the AI Disclosure Statement (docs/legal/ai-disclosure-statement.md, public: https://ataski.com/legal/ai-disclosure).

This document is not legal advice. It reflects Ataski's operating practices for the benefit of customers, prospects, and the users whose data our customers process through the Service.

1. Who we are and what this policy covers

Ataski operates ataski.com and the Ataski platform — a subscription catalogue of AI workers ("AI employees") that perform business tasks for our customers, including sales outreach, customer support, renewals, meeting coordination, and financial reporting. Each AI worker runs under a defined capability matrix and is supervised for quality.

This Privacy Policy explains:

What personal data we collect and process, and in what role (controller vs. processor / Service Provider).
How we handle data we receive from Google APIs and Microsoft Graph, including the limited-use commitments those platforms require (see §6 and §7 — read these first if you are evaluating our Google or Microsoft connectors).
Who our sub-processors are.
How we handle data with large language models (LLMs).
How long we keep data and how to exercise your rights.

Our service is business-to-business. It is not directed to, and we do not knowingly collect personal data from, children (see §16).

2. Controller vs. processor — which hat we wear

Ataski processes personal data in two distinct capacities. Knowing which applies is essential to understanding your rights and ours.

2.1 Ataski as processor / Service Provider (the common case)

For the personal data our customers route through the Service — their prospects, leads, end-customers, contract counterparties, meeting participants, and the contents of emails, documents, and records they ask an AI worker to act on — the customer is the controller and Ataski is the processor (GDPR Article 28) / Service Provider (CCPA/CPRA §1798.140(ag)). We process this data only on the customer's documented instructions, which are the Master Services Agreement, this Privacy Policy, the Data Processing Addendum, and the configuration the customer sets in the Ataski dashboard (campaign targeting, mailbox/calendar connections, capability-matrix scope, PII-handling mode).

As a Service Provider, Ataski is not a data broker under CCPA §1798.140 or any state Data Broker Registration law. We do not aggregate, sell, or charge for access to contact personal information. Contact records stay within each customer's Row-Level-Security (RLS) tenant scope. Only non-personal firmographic information (company name, industry, public business-registry facts) may be cached across tenants.

If you are a data subject of one of our customers' campaigns (for example, you received outreach), the customer is your first point of contact as controller. Ataski will assist that customer, and we also operate a direct opt-out/erasure path described in §13.4.

2.2 Ataski as controller

For the personal data we collect to operate our own business — your account and identity information, billing details, support correspondence with us, website analytics, and security/operational logs — Ataski is the controller. This policy is your direct notice for that processing.

3. The personal data we collect

3.1 Account and identity data (Ataski as controller)

Name, work email, and organisation affiliation, captured at sign-up and authentication via our auth provider (WorkOS).
Role assignment within your workspace (e.g., founder, finance_lead, operator).
Authentication events and session metadata.

3.2 Customer-connected and customer-uploaded business data (Ataski as processor)

Depending on which AI workers a customer enables, this can include:

Connected mailbox data — when a customer connects Gmail or Outlook, the data exposed by the specific scopes granted (see §6 and §7). We send outreach as the customer's own connected mailbox (bring-your-own-mailbox); we never send a customer's outreach from a shared Ataski domain.
Connected calendar data — free/busy availability and, where the customer enables meeting booking, confirmed events written to their calendar (see §6 and §7).
CRM and accounting data — read-only deal, contact, company, and financial data from connected systems (for example, HubSpot, Salesforce, Pipedrive, Zoho, QuickBooks, Xero, and Stripe), accessed under the customer's own OAuth grant or API key.
Uploaded documents — contracts and other files a customer uploads or connects (for example, via Google Drive, OneDrive, or DocuSign) for roles such as AI Contract Intelligence.
Meeting content — transcripts the customer uploads, or (when re-enabled with per-participant consent) live-meeting audio/transcripts captured by a meeting bot. See the AI Disclosure Statement for the consent and recording-disclosure rules.

3.3 Contact data processed inside the AI workers (Ataski as processor)

The recipient/prospect data a customer's roles act on — names, work email addresses, job titles, employer, public professional URLs, and engagement signals — held strictly within the customer's RLS tenant scope. Where a customer's campaign requires identifying additional prospects, Ataski uses a compose pattern (public web search + deliverability verification) rather than buying contact lists from email-finder/data-broker vendors. See the Sub-Processors disclosure for the exact data vendors and what each sees.

3.4 Usage, billing, and operational data

Billing — subscription status and payment details handled by Stripe; Ataski does not store full card numbers.
Product analytics — page-view and event telemetry (PostHog), used to improve the product; configured to avoid sending personal data where possible.
Logs and telemetry — structured application logs (correlation id, tenant id, task id, model used, cost, latency), an append-only audit log of every action, error traces (Bugsink, with PII off by default), LLM traces (Langfuse), and uptime metadata (Better Stack).

3.5 Special categories

By design, Ataski does not seek special-category data (GDPR Article 9). Every role's capability matrix refuses medical, biometric, genetic, religious, political, sexual-orientation, or trade-union data on intake. For AI Contract Intelligence, a deterministic preflight scanner can detect and (per the customer's selected mode) redact or block high-sensitivity categories before any document content is sent to an LLM sub-processor — see §9 and the Data Processing Addendum.

4. How and why we use personal data (purposes and legal bases)

We use personal data only for the purposes for which it was provided (purpose limitation), and we collect only what each role needs (data minimisation).

Purpose	Capacity	GDPR/UK GDPR legal basis (where applicable)
Providing the AI-worker features the customer enabled (draft/send outreach, read availability, book meetings, classify inbound, extract document data, generate financial narratives)	Processor	Customer's lawful basis as controller; Ataski acts on Art. 28 instructions
Creating and securing customer accounts; authentication	Controller	Contract (Art. 6(1)(b)); legitimate interests in security (Art. 6(1)(f))
Billing and tax	Controller	Contract; legal obligation (Art. 6(1)(c))
Product analytics, debugging, and service improvement	Controller	Legitimate interests (Art. 6(1)(f))
Security, abuse prevention, audit, and incident response	Controller / Processor	Legitimate interests; legal obligation
Responding to data-subject and consumer requests	Controller / Processor	Legal obligation

We do not use personal data for automated decisions producing legal or similarly significant effects on individuals without the customer's controller-side determination.

5. AI / LLM handling

Ataski runs a two-model LLM stack: an Anthropic Claude model as the worker and an OpenAI GPT-5 model as a cross-family supervisor, plus OpenAI embeddings (text-embedding-3-small). We do not use Google Gemini or any Google LLM, and no customer data is routed to Google for model inference. When the two models disagree on a high-stakes task, the task is routed to an internal operator inbox — there is no third-model tiebreaker.

Key commitments:

We process LLM inputs/outputs only to provide the feature the customer enabled.
Our LLM providers operate under enterprise terms with a zero-retention / no-training posture: prompts and completions are not used to train their models and are deleted on a short retention window. See the Sub-Processors disclosure for the per-provider statement.
Ataski does not use customer personal data — and specifically does not use any data obtained through Google APIs or Microsoft Graph — to develop, improve, or train generalised or non-personalised AI/ML models. Any improvement of our own extractor evaluation suites uses tenant-scoped feedback that is stripped of tenant/user identifiers before promotion, reviewed individually (see §9).
Enterprise customers may bring their own Anthropic/OpenAI API key; in that case their LLM calls run through their key and Ataski's invoice excludes that LLM cost.

6. Google API Services — User Data and Limited Use

This section governs all data Ataski accesses through Google APIs (Gmail, Google Calendar, Google Drive, Google Sheets).

Ataski's use and transfer of information received from Google APIs to any other app will adhere to the Google API Services User Data Policy, including the Limited Use requirements.

6.1 Scopes we request and the reason for each

We request only the narrowest scopes needed for a feature you explicitly enable, and only after you grant consent through Google's OAuth screen. Access is per-connection and can be revoked at any time from your Google Account.

Google scope	Why we request it	What it lets us do
`https://www.googleapis.com/auth/gmail.send`	Send the outreach you instruct, from your own connected Gmail mailbox (bring-your-own-mailbox).	Send a message you (or your supervised AI worker) have prepared. This scope is send-only; it does not grant any ability to read your mailbox.
`https://www.googleapis.com/auth/gmail.readonly` and/or `https://www.googleapis.com/auth/gmail.metadata`	Deduplicate prospects already in your inbox so we don't double-contact them — only when you enable inbox dedup.	We query thread metadata (thread ids and From/To/Subject headers) to detect an existing conversation. We do not read message bodies for this feature.
`https://www.googleapis.com/auth/gmail.compose` and/or `https://www.googleapis.com/auth/gmail.modify`	Create a draft in your Gmail Drafts folder when you choose "save to drafts" instead of auto-send.	Create a draft you then review and send yourself.
`https://www.googleapis.com/auth/calendar.freebusy`	Read your availability to propose meeting times in outreach.	Read busy/free intervals only — never event titles, attendees, or descriptions.
`https://www.googleapis.com/auth/calendar.events`	Write a confirmed meeting onto your calendar — only when you enable meeting booking.	Create the agreed event with the prospect as attendee. This feature ships disabled until the verified scope is live.
`https://www.googleapis.com/auth/drive.readonly`	Ingest the contract/financial documents you point us to, for the document-reading roles.	List and download the specific files you authorise. We do not modify your Drive.
`https://www.googleapis.com/auth/spreadsheets`	Push the AI Financial Analyst's output into a Google Sheet you already own and specify.	Write model output into the sheet id you supply. We do not create or browse your other sheets.

6.2 What Google user data we access

The Gmail message we send on your instruction (and the draft we create where enabled); Gmail thread metadata for dedup where enabled; calendar free/busy intervals and, where you enable booking, the single confirmed event we write; the specific Drive files you authorise; and the contents of the specific Google Sheet you designate.

6.3 Limited Use commitments

For all data received from Google APIs, Ataski commits that we:

Use it only to provide or improve the user-facing features you have enabled.
Do not transfer it to others except (a) as necessary to provide or improve those user-facing features, (b) with your consent, (c) for security purposes (e.g., investigating abuse), (d) to comply with applicable law, or (e) as part of a merger, acquisition, or sale of assets, with notice to you.
Do not use it to develop, improve, or train generalised or non-personalised AI and/or ML models. We do not send Google user data to any LLM for model training, and our worker/supervisor LLM providers operate under no-training terms.
Do not sell it, and do not use or transfer it for serving advertisements.
Do not allow humans to read Google user data, except: (a) with your affirmative agreement for specific messages; (b) where necessary for security purposes (such as investigating a bug or abuse); (c) to comply with applicable law; or (d) where the data has been aggregated and anonymised and is used solely for internal operations. Routine processing is automated.

6.4 Verification and revocation

The Gmail and Calendar write scopes are "restricted"/"sensitive" scopes subject to Google's verification and, where applicable, security assessment. Until verification is complete, restricted features are limited or disabled. You can review and revoke Ataski's access at any time at Google Account → Security → Third-party access. Revocation stops further access; data already processed to deliver a feature is retained and deleted per §12.

7. Microsoft Graph — equivalent limited-use posture

When you connect Microsoft 365 / Outlook, Ataski accesses Microsoft Graph under a single Ataski-registered application, requesting only:

Microsoft Graph permission	Why we request it
`Mail.Send`	Send the outreach you instruct, from your own connected Outlook mailbox.
`Calendars.Read`	Read your availability (busy/free) to propose meeting times.
`Calendars.ReadWrite`	Write a confirmed meeting to your calendar — only when you enable meeting booking.
`Files.Read.All`	Ingest the OneDrive documents you point us to, for the document-reading roles.

For all data received through Microsoft Graph, Ataski applies the same limited-use posture described in §6.3: we use it only to provide the feature you enabled; we do not sell it; we do not use it to train generalised/non-personalised AI/ML models; we do not allow humans to read it except with your consent, for security, to comply with law, or in aggregated/anonymised form for internal operations; and we transfer it only as needed to provide the feature, with your consent, for security/legal reasons, or in a merger with notice. Our use complies with the Microsoft APIs Terms of Use and Microsoft's data-handling requirements. You may revoke Ataski's access from your Microsoft account at any time.

8. Sub-processors

Ataski engages a limited set of third parties to process Customer Data on our behalf. The canonical, current list — with each vendor's legal entity, purpose, data categories accessed, location, and DPA link — is published at https://ataski.com/legal/subprocessors and in docs/legal/subprocessors.md, generated from the machine-readable source of truth in src/ataski/legal/subprocessors.py.

By category, our sub-processors are:

Hosting & infrastructure / database — Hetzner (compute), Neon (Postgres), Cloudflare (DNS/CDN/WAF and R2 object storage).
Authentication — WorkOS.
Email delivery — Wildbit/Postmark (transactional + inbound).
LLM / AI — Anthropic (Claude worker) and OpenAI (GPT supervisor + embeddings). No Google/Gemini.
Meetings — Recall.ai (meeting bot identity + transcription).
Contact-data providers (PAYG only) — Bright Data, Perplexity, ZeroBounce, Apify.
Billing — Stripe.
Observability & operations — Langfuse, Bugsink, PostHog, Better Stack.

Customer bring-your-own connectors (Gmail, Outlook, Salesforce, HubSpot, Pipedrive, Slack, QuickBooks/Xero, Apollo, NIPR, etc.) are accessed only under the customer's own credential. Ataski is not a sub-processor for those — the customer's relationship with each is direct.

Customers receive 30 days' advance notice (in-app banner + email to the workspace admin) before a new sub-processor goes live, and may object during that window, as described in the DPA and the Sub-Processors page.

9. AI Contract Intelligence — document PII handling

For the AI Contract Intelligence role, a deterministic preflight scanner runs locally on Ataski's host (no external service) and detects 12 categories of personal data in uploaded contracts. Each customer selects a per-tenant PII-handling mode at https://app.ataski.com/app/settings#pii-policy:

detect_only (default) — the document text is sent to the LLM sub-processor; detected categories are recorded on the append-only extraction audit row.
redact_selected — selected categories are replaced with a visible [REDACTED-<category>] sentinel before any text leaves Ataski for the LLM.
block_high_sensitivity — if SSN, credit-card, bank-routing, health-condition, or biometric data is detected, the extraction refuses and no document bytes leave Ataski for any sub-processor.

Scanned image-only PDFs are OCR'd locally (open-source Tesseract on the Ataski host); image bytes never leave Ataski — only the OCR'd text follows the selected mode. Full detail, including the closed category set and customer-instruction status, is in the Data Processing Addendum and the Data Retention Policy.

Customer corrections submitted via the dashboard feedback control are stored in the customer's RLS scope and used to improve our extractor evaluation suite; any correction promoted into a regression test is stripped of tenant/user identifiers first and reviewed individually, and corrections are deleted on the standard offboarding schedule (§12).

10. AI disclosure in customer communications

Ataski never hides that a recipient is interacting with an AI-assisted system. AI-drafted outreach is reviewed under the customer's supervision and carries disclosure consistent with applicable law (e.g., California SB 1001 bot disclosure and CCPA notice-at-collection for California recipients), plus one-click unsubscribe and a suppression mechanism. Meeting recording, when enabled, carries participant-consent capture and an AI-presence announcement. The controlling rules are in the AI Disclosure Statement.

11. Security and multi-tenant isolation

Tenant isolation — every customer-data table is protected by Postgres Row-Level Security; queries are automatically scoped to the requesting tenant. Cross-tenant aggregation of personal data is prohibited by code path.
Encryption — data is encrypted in transit (TLS 1.2+) and at rest (Neon-managed AES-256 for the database; server-side encryption for R2 file storage). Connected-account OAuth tokens are stored encrypted.
Access control & audit — least-privilege access, SSO via WorkOS, and an append-only audit log of every administrative action, LLM call, and external API/data-vendor call, retained per the Data Retention Policy.
Cost guardrails — per-tenant daily caps and global kill switches limit blast radius from a compromised credential or runaway loop.

A current security overview is published at https://ataski.com/legal/security. Personal-data breaches are notified to affected customers without undue delay and within 72 hours of Ataski becoming aware, per the DPA.

12. Data retention and deletion

We keep personal data only as long as needed for the purpose it was collected, then delete it. Full periods are in the Data Retention Policy. Key timelines:

Customer Data (CRM, outreach history, drafts, documents, transcripts, contact records) is retained while the subscription is active and automatically deleted 30 days after offboarding. During that 30-day window the data remains exportable via /api/export (admin: /app/settings/export.json); a reminder is sent at day 7 and day 23, and an automated sweep deletes all tenant-scoped rows at day 30.
Suppression lists are retained indefinitely as a legal obligation (CAN-SPAM / CCPA) so a suppressed contact is never re-contacted.
Audit log and external-call (MCP) logs are retained 24 months for regulatory evidence, then purged.
Right-to-delete request logs are retained 7 years as compliance evidence.
Firmographic cache (non-personal) is deleted 30 days after last access; rows referenced only by a deleted tenant are removed in the offboarding sweep.

13. Your rights

You have the right to access, rectify, erase, restrict, port, and object to processing of your personal data, and to withdraw consent where processing relies on it. Where Ataski is the controller (your account data), contact us per §17. Where Ataski is a processor (data our customers process through the Service), please contact the relevant customer (controller); Ataski will assist them per the DPA. You also have the right to lodge a complaint with your supervisory authority.

13.2 California (CCPA/CPRA) and other US state laws

California residents (and residents of states with comparable laws) have the right to know/access, delete, correct, and opt out of the sale or sharing of personal information, and the right to non-discrimination for exercising these rights.

Ataski does not "sell" personal information and does not "share" it for cross-context behavioral advertising, as those terms are defined under the CCPA/CPRA. Accordingly, Ataski acts only as a Service Provider for Customer Data and is not a data broker.

13.3 How to exercise rights against Ataski (controller data)

Email privacy@ataski.com. We verify requests and respond within the statutory timeline (generally 45 days under CCPA, 30 days under GDPR; extensions where permitted, with notice). Tenant administrators can also export their workspace data at any time via /app/settings/export.json.

13.4 Direct opt-out / erasure for outreach recipients

If you received outreach sent through Ataski on a customer's behalf and want your data removed, use the one-click unsubscribe in the message or the link to /privacy/optout. Opt-outs propagate to all of that customer's roles, are added to the suppression list to prevent re-contact, and are propagated to relevant sub-processors within the timelines in the DPA.

14. International data transfers

Ataski is operated from, and primarily uses sub-processors located in or operating from, the United States and the EU (see the Sub-Processors disclosure for each entity's location). For transfers of personal data from the EEA, UK, or Switzerland to a third country, we rely on the European Commission's Standard Contractual Clauses (and the UK International Data Transfer Addendum), and on the EU-US Data Privacy Framework where the receiving sub-processor is certified. Ataski maintains a Transfer Impact Assessment covering the US destination and the listed sub-processors; a current summary is available to customers on request via legal@ataski.com. Details are in the DPA.

15. Cookies and similar technologies

Ataski uses a minimal set of cookies/local storage strictly necessary to authenticate sessions and protect the service, plus limited first-party product analytics (PostHog) to understand and improve usage. We do not use advertising or cross-site tracking cookies. Where required, we obtain consent for non-essential analytics. You can control cookies through your browser settings.

16. Children

The Service is a business tool intended for use by organisations and their personnel. It is not directed to children, and we do not knowingly collect personal data from anyone under 16 (or under 13 where that is the applicable threshold). If we learn we have collected such data, we will delete it.

17. Contact

Privacy questions / data-subject & consumer requests: privacy@ataski.com
Legal, DPA, and transfer documentation: legal@ataski.com
Security: see https://ataski.com/legal/security

Ataski will designate an EU/UK Article 27 representative and (where required) a Data Protection Officer; contact details will be added here on appointment. Until then, all privacy enquiries are handled via the addresses above.

18. Changes to this policy

We may update this Privacy Policy to reflect changes to the Service, our sub-processors, or the law. Material changes are notified to customers via in-app banner and email to the workspace admin. New sub-processors carry the 30-day advance notice described in §8. The "Last updated" date at the top reflects the current version; superseded versions are retained internally for audit.