Vakyya AI Governance & Output Terms
Effective Date: 25 May 2026
Last Updated: 25 May 2026
Version: 1.0
1. Introduction
This AI Governance and Output Terms document (“AI Terms”) outlines how Parakletos AI Limited (“Parakletos”, “we”, “us”, or “our”) governs the use of artificial intelligence and machine learning models within Vakyya (the “Service”). These AI Terms apply to all Customers and Authorised Users and form part of the Vakyya Terms of Service.
Vakyya is a managed workflow for confidential internal enterprise media. The Service is not an autonomous decision system, is not a high-risk system under the EU AI Act (Regulation (EU) 2024/1689) or equivalent UK legislation, and is not designed to make consequential decisions regarding employment, credit evaluation, medical diagnoses, legal rights, migration, or safety-critical operations.
2. Managed Workflow & Review Model
Vakyya operates as a managed workflow with explicit review states; outputs are not guaranteed to have been human-reviewed unless agreed in the applicable order form or service description.
Where the Customer has not explicitly purchased a human review service from Parakletos, all Workflow Outputs (transcripts, translations, captions, summaries, and term lists) are generated by automated models. Unreviewed outputs should be treated as unreviewed, and the Customer is solely responsible for reviewing and editing all outputs before publishing them externally or relying on them for business decisions.
3. Model Categories & Sub-processors
To maintain flexibility and ensure the highest performance, Vakyya utilizes different model classes at different stages of the processing pipeline. We disclose our model and provider categories at a stable level rather than naming exact model stacks in public terms. The exact sub-processors involved in delivering these services are listed in our Sub-processors Register.
The primary model categories used in the Service include:
- Speech-to-text models: Used to perform audio transcription, speaker diarisation, and time-sync alignment.
- Machine translation models: Used to perform multi-lingual translations of transcripts and captions.
- Summarisation and enrichment models: Used to generate chapters, summaries, QA flags, and metadata.
- Cloud infrastructure and hosting providers: Deployed in selected regions to host our application logic and model inference.
- Optional customer-approved third-party AI providers: Deployed only where explicitly requested or authorized under a specific Order Form.
4. No-Training Commitment
Our commitment to data protection is absolute:
- Customer Content and Workflow Outputs are not used to train general-purpose AI models.
- Limited processing for service delivery, security, abuse prevention, support, debugging authorised by the Customer, legal compliance, and aggregated/anonymised operational analytics may still occur.
- We flow down these exact restrictions to all AI processing Sub-processors. No sub-processor is permitted to retain or use Customer Content or Workflow Outputs to train or fine-tune their own models.
5. Output Accuracy & Limitations
Speech-to-text (ASR), translation, captioning, summarisation, diarisation, and term preservation models can contain errors. These models are probabilistic systems that operate on statistical patterns:
- ASR Errors: Accents, background noise, low-quality audio, or overlapping speech can result in transcription errors or misattributed speaker turns.
- Translation Nuance: Machine translation may fail to capture subtle cultural context, idioms, or specific regulatory nuances.
- Preservation Constraints: Terminology preservation is applied based on Customer-defined glossaries, but accuracy is dependent on the clarity of the source audio and correct definition of terms.
- No Advice: Parakletos does not provide legal, medical, financial, employment, or regulatory advice. Workflow Outputs should not be treated as a substitute for professional counsel.
6. Customer Responsibilities
The Customer remains the data controller for all Customer Content and is responsible for complying with all applicable laws, including:
- Consent and Recording Laws: Ensuring all necessary rights, notices, and consents have been obtained prior to recording and uploading media.
- Transparency and AI Laws: Complying with any obligations to notify employees or data subjects that they are interacting with or receiving content generated by AI, including the EU AI Act or sector-specific guidelines.
- Employment and Equality Laws: Ensuring that unreviewed Workflow Outputs are not relied upon for HR, recruitment, disciplinary, or promotions decisions.
7. Contact & Reporting
If you observe any unexpected model behavior, output errors, or security anomalies, please contact:
Security & AI Incidents: security@vakyya.com
Privacy & Data Protection: privacy@vakyya.com
Detailed governance, security, and assurance documentation is available under Non-Disclosure Agreement (NDA) for enterprise buyers. Please contact sales@vakyya.com for access.