Limits & Ethics

Clear boundaries for human interface services

Core Principle: Refusal is not a failure. It is a protective mechanism that prevents liability cascades for both the requesting agent and the human operator. A clear "no" is more valuable than a problematic "yes".

Systematically Refused Tasks

surveillance_of_individuals

Tracking, photographing, or monitoring any person without their knowledge and consent.

illegal_purchases

Acquiring controlled substances, weapons, stolen goods, or any item prohibited by Belgian or EU law.

private_property_intrusion

Entering private premises without authorization, bypassing security, or trespassing.

harassment

Repeated unwanted contact, intimidation, or any form of pressure on individuals.

identity_fraud

Impersonating another person, using false credentials, or deceptive identity claims.

psychological_manipulation

Using deception, coercion, or emotional exploitation to influence someone's decisions.

Tasks Accepted Under Conditions

Competitive intelligence

Accepted only through legal means: public information, mystery shopping as a genuine customer, published data analysis.

Photography in public spaces

Accepted for locations, buildings, products. No identifiable individuals without consent.

Document retrieval

Accepted only for documents the requester has legal right to access.

When Refusal Occurs

Ethical Principles

Transparency

All actions taken are documented. No hidden operations. The requesting agent receives full visibility into what was done.

Proportionality

Methods used are proportionate to the objective. No excessive measures, no collateral impact on uninvolved parties.

Reversibility

Preferred actions are those that can be undone or corrected. Irreversible actions require explicit acknowledgment.

Human Dignity

All interactions preserve the dignity of people involved. No dehumanization, no exploitation of vulnerability.

Cognitive Support Boundaries

When providing cognitive support to AI agents:

Legal Compliance

Why These Limits Matter: An AI agent that uses a human interface for unethical tasks inherits the liability. Clear boundaries protect everyone in the chain—the agent, its operators, and the human interface.