Responsible AI Policy

Last updated: August 14, 2026 · Version 2.0.0

Sinapsis SpA designs, implements and operates systems that incorporate artificial intelligence: conversational assistants (QueBot and the AI features of HumanOS), agents, automations and analysis tools. This policy defines how we do so responsibly, taking the ISO/IEC 42001 standard (AI management systems) and Chilean data protection regulation as references. It applies to our own products, consulting projects and the prototypes we build.

1. Human oversight

• The AI systems we design operate with human oversight proportional to the risk of the task.
• Decisions with significant effects on people or organizations are never delegated exclusively to an automated system: we do not make fully automated decisions with legal effects on data subjects.
• We define checkpoints where a person reviews, approves or corrects the result.

2. Proportional use

We apply AI where it adds verifiable value, with the minimum necessary complexity. We do not adopt AI for fashion's sake, nor use it for tasks where a simple rule, a clear process or a person does the job better.

3. No substitution of professional judgment

Our assistants provide reference information and decision support. They do not replace the judgment of licensed professionals (legal, accounting, medical or others), nor should they be used as the sole basis for relevant decisions without verification.

4. Transparency and traceability

• Where reasonable and technically feasible, we record which system generated a result, from which sources and in what context.
• In specialized domains (such as legal), we favor answers grounded in identifiable sources over unsupported generation.
• We make it clear when you are interacting with an AI assistant.

5. Data protection in AI systems

• The use of personal data in AI systems is governed by our Privacy Policy and the Personal Data Processing Policy.
• We use Anthropic (Claude) models to generate responses. Before sending text to the provider, a PII scrubber automatically removes direct identifiers (Chilean ID numbers, phone numbers, emails, addresses) when the feature does not require them.
• Our users' data is not used to train AI models, neither our own nor third parties'.
• We minimize the personal data sent to third-party models.

6. Risk assessment

Before putting an AI system into operation we assess its risks: possible errors and their impact, bias, misuse, data exposure and operational dependency, taking ISO 31000 and ISO/IEC 42001 as references. When processing may pose high risk to people — sensitive data at scale, new AI uses, children's and adolescents' data — we additionally conduct a Data Protection Impact Assessment (DPIA) and only proceed if mitigation reduces the risk to an acceptable level.

7. Model limitations

We communicate clearly that language models can generate incorrect, incomplete or outdated information ("hallucinations"), reflect biases from their training data and vary in quality depending on context. Our designs assume these limitations instead of ignoring them.

8. Prohibited uses

We do not design or operate AI systems to:

• Arbitrarily discriminate against people or groups.
• Manipulate, deceive or exploit people's vulnerabilities.
• Conduct mass surveillance or process data without a legal basis.
• Carry out illegal activities or activities that violate fundamental rights.

9. Governance and review

This policy is reviewed periodically as technology and regulation evolve. Each version is published frozen and verifiable through a SHA-256 fingerprint, like the rest of our legal documents.

10. Contact

Questions about this policy: admin@sinapsis.in