1. Purpose and Scope
This policy governs how Elementyl Intelligence designs, deploys, and manages AI systems across our internal operations and in the solutions we build for clients. It applies to all AI tools, voice agents, chatbots, and agentic workflows whether built in-house or delivered through our technology partners.
It is written for two audiences: our team, collaborators, and partners who need clear operating standards, and our clients, who deserve to know exactly how we think about the technology we bring into their businesses.
AI should reduce friction, expand capability, and free up human attention for the work that actually matters. When it does the opposite, we have not done our job.
2. Core Principles
Every AI system we design or deploy is evaluated against these principles. They are not aspirational. They are operational requirements.
Human Amplification Over Replacement
We build AI that makes the people already in the business more capable. Existing employees are to be upskilled or redeployed, not eliminated. We do not design systems whose primary purpose is to remove people from a workforce, and we do not accept engagements where that is the stated goal.
Transparency and No Dark Patterns
Clients and their customers should know when they are interacting with an AI system. We do not deploy agents or chatbots designed to pass as human without explicit informed consent. We reject dark patterns in AI design: systems built to obscure their nature, manipulate user behavior, or exploit cognitive biases have no place in our work.
Data and Privacy Honesty
We are forthcoming about how data is collected, used, and stored in every system we build. We collect only what is functionally necessary. Client and end-user data is never used to train models without explicit written consent. We disclose all third-party platforms involved in data processing before a project begins.
Active Bias Mitigation
We actively work to identify and counter bias in AI systems across race, gender, age, and any other dimension. This is not passive. We treat bias testing as a formal step in every build, and discriminatory output patterns are a hard blocker on deployment, not a post-launch issue.
Shared Accountability
When an AI system causes harm or behaves incorrectly, responsibility is shared. We are accountable for how the system was built and what safeguards were designed in. Clients are accountable for how the system is operated and maintained after handoff. This division is documented and signed before any deployment goes live.
Reversibility
Clients retain the right to modify or decommission any AI system we build for them. We design for clean exits and document everything so no client is ever locked in by complexity we created.
3. Client Engagement Standards
3.1 Values Alignment Before Engagement
Before agreeing to any engagement, we assess whether the client's goals are compatible with our operating principles. This is a qualifying step, not a formality.
We counsel clients on the realistic capabilities and limitations of AI, including cases where their initial assumptions about what AI can or should do need to be corrected. A client who wants efficiency gains is a client we can work with. A client whose primary stated goal is maximizing workforce elimination is not.
Intent matters. We reserve the right to decline any engagement where the client's purpose conflicts with our commitment to people.
3.2 Before We Build
- We confirm that AI is actually the right tool for the problem before development begins.
- We document the intended use case, expected behavior, and success criteria in writing.
- We flag any use cases we consider outside the bounds of ethical deployment and explain why.
3.3 During Development
- We test outputs against real-world scenarios including edge cases and adversarial inputs before handoff.
- We document how the system works in plain language, not just technical specifications.
- We identify all data sources the system depends on and flag concerns about quality or provenance.
- We conduct bias testing as a formal step in the build process.
3.4 Deployment Requirements
A system does not go live until the following conditions are met:
- The client has reviewed and signed off on the safety and maintenance requirements.
- The client has demonstrated sufficient understanding of those requirements to maintain them.
- A named human owner at the client organization has accepted operational accountability.
Clients who are unwilling to engage with safety maintenance are not cleared for deployment. This is a hard requirement.
3.5 After Deployment
- We provide a monitoring framework so clients can track system performance over time.
- We establish a clear process for reporting issues and requesting changes.
- We conduct a post-deployment review within 30 days of go-live for all voice agent and agentic workflow deployments.
4. Accountability When Things Go Wrong
AI systems make mistakes. What matters is how responsibility is structured and how errors are corrected.
4.1 Division of Responsibility
Elementyl Intelligence is responsible for:
- The design and architecture of the system.
- The safeguards built in at launch.
- The accuracy of documentation and training provided to the client.
- Bias testing and pre-deployment review.
The client is responsible for:
- Operating the system according to the guidelines provided.
- Maintaining the safety protocols they agreed to at deployment.
- Notifying Elementyl Intelligence promptly when something behaves unexpectedly.
This division is formalized in a signed agreement before go-live. Clients who have not followed documented maintenance requirements cannot hold Elementyl Intelligence solely liable for resulting errors.
4.2 Incident Response
- We notify affected clients within 48 hours of identifying an incident.
- We provide a written incident summary within 5 business days.
- We work with the client to remediate and document corrective actions.
5. Prohibited Uses
We will not design, build, or deploy AI systems for the following purposes regardless of client request or commercial incentive:
- Surveillance of employees without their knowledge and consent.
- Generation of misinformation, manipulated media, or deceptive content intended to mislead.
- Automated decisions with material consequences to individuals, such as hiring or lending, without human review at the point of decision.
- Collection or processing of data belonging to minors.
- Any use that would require violating applicable privacy law including CCPA or GDPR where applicable.
- Impersonation of a real person or organization without authorization.
- Dark patterns designed to manipulate, deceive, or exploit end users.
If a client request falls into a gray area, we will say so directly and work through it transparently rather than proceeding without disclosure.
6. Data Governance
6.1 Data Handling Standards
- Data is classified at intake as public, internal, confidential, or restricted.
- Confidential and restricted data is never stored in prompt histories, testing environments, or shared workspaces without encryption.
- Client business data is not used for internal benchmarking or product development without a signed data use agreement.
6.2 Third-Party Platforms
Our solutions may use third-party AI platforms and voice infrastructure. Where we rely on these platforms we:
- Review their data processing terms before incorporating them into a client solution.
- Disclose which platforms are involved in writing before project kickoff.
- Flag any material changes to third-party terms that could affect data handling.
7. Internal AI Use
The same principles that govern our client work apply to how we use AI internally.
- We do not input confidential client information into public-facing AI tools without assessing those tools' data handling policies.
- AI-generated content used in client-facing materials is reviewed by a human before delivery or publication.
- We maintain a list of approved internal AI tools and review it quarterly.
- Team members are expected to disclose AI involvement in work products when that involvement is material to quality or reliability.
8. Review and Accountability
This policy is reviewed at minimum annually and updated when meaningful changes occur in the business or the technology landscape. It is a working document.
8.1 Policy Owner
This policy is owned by the Founder and CEO of Elementyl Intelligence. Questions, concerns, or suggested revisions: anne@elementylintelligence.ai
8.2 Grievance Process
If a client, partner, or employee believes this policy has been violated, they can submit a written description to anne@elementylintelligence.ai. We commit to acknowledging receipt within 2 business days and providing a substantive response within 10 business days.
8.3 Version History
| Version | Date | Notes |
|---|---|---|
| 1.0 | March 2026 | Initial release. |
"We hold ourselves to these standards because our clients are trusting us to bring a powerful technology into their businesses responsibly. That trust is not abstract. It is the foundation of everything we do."