These AI Terms supplement the Terms of Service and govern your use of artificial intelligence features within Anyrow. If a conflict exists, these AI Terms prevail for matters concerning AI processing.
1. Scope
These terms apply whenever Anyrow uses AI models to process your documents, extract structured data, or generate outputs on your behalf. The AI features are integral to the service described in the Terms of Service.
2. AI Models and Subprocessors
Anyrow uses third-party AI models to perform document extraction. The authoritative list of AI model providers, their processing locations, and transfer mechanisms is maintained at /legal/subprocessors.
Each provider operates under a contractual agreement that prohibits them from using your inputs or outputs to train or improve their models. We use only paid API tiers where such protections are contractually guaranteed.
3. No Training on Your Data
Anyrow does not use your documents, extracted data, or corrections to train foundation AI models — neither our own nor those of our subprocessors.
Specifically:
- Your uploaded documents are sent to AI model providers solely for the inference call that serves your extraction request
- AI providers receive your data under API terms that contractually prohibit training on customer inputs
- We do not aggregate customer data across accounts for model training purposes
- Extracted outputs stored in your workspace are not fed back into any training pipeline
4. Adaptive Extraction (Future Capability)
Anyrow may introduce workspace-level extraction improvements that learn from your corrections to increase accuracy over time. If and when this capability is deployed:
- It will be opt-in — disabled by default, activated only by explicit workspace administrator action
- Learned adaptations will be tenant-isolated — your corrections never improve extraction for other customers unless you explicitly contribute to a shared template
- Shared template contributions, if offered, will be separately opt-in and anonymised before aggregation
- You may delete learned adaptations at any time; deletion removes them from active use within 30 days
We will update these AI Terms and notify you by email at least 30 days before any adaptive extraction feature becomes available.
5. Input and Output
5.1 Ownership
Your Input (uploaded documents) and Output (extracted data, structured rows) remain your property. The limited processing license you grant under the Terms of Service applies.
5.2 Responsibility
You are responsible for:
- Ensuring you have lawful authority to submit your documents for AI processing
- Reviewing extraction outputs before relying on them for decisions
- Not submitting documents that contain content prohibited under the Acceptable Use section of the Terms of Service
6. Accuracy and Limitations
AI extraction is not error-free. Outputs are generated by artificial intelligence and may contain inaccuracies, omissions, or misinterpretations. Anyrow makes no warranty as to the accuracy, completeness, or reliability of AI-generated outputs.
You must:
- Review outputs before acting on them, especially for financial, legal, medical, or compliance-critical data
- Not represent AI-generated outputs as verified, audited, or human-reviewed unless you have independently verified them
- Maintain your own records where regulatory requirements apply
Extraction accuracy varies by document type, quality, language, and layout complexity. Similar documents may produce different results across extraction runs.
7. Data Retention by AI Providers
AI model providers process your documents transiently for the duration of the inference call. Under our contractual arrangements:
- Providers do not retain your input data after the inference response is returned
- Providers may retain limited metadata (token counts, error logs) for abuse prevention and service reliability, as documented in their respective DPAs linked from our Subprocessors page
- Anyrow itself retains your documents and extracted data in your workspace until you delete them, subject to the retention terms in the Privacy Policy
8. Prohibited High-Risk Uses
In addition to the restrictions in the Terms of Service, you must not use Anyrow AI features as the sole basis for automated decisions in the following categories without appropriate human oversight:
- Credit scoring, lending, or insurance underwriting
- Employment decisions (hiring, termination, performance evaluation)
- Criminal justice risk assessment
- Determination of access to essential services (housing, healthcare, education)
- Biometric identification or categorisation of natural persons
- Any use classified as "high-risk" under the EU AI Act (Regulation (EU) 2024/1689), Annex III
You may use Anyrow to extract data that informs these decisions, provided a qualified human reviews and independently validates the output before any consequential action is taken.
9. Transparency
In accordance with the EU AI Act (Art. 50) and as general good practice:
- Anyrow clearly identifies AI-generated outputs in the product interface
- Extraction results are presented as AI-generated and editable, not as verified facts
- You must disclose to your own end users, where applicable, that data was extracted using AI
10. Changes to These Terms
We may update these AI Terms to reflect changes in AI providers, capabilities, or applicable law. Material changes will be notified by email at least 30 days before they take effect. The most recent update date is shown at the top of this page.