MODEL GOVERNANCE
Who decides about Na'at and how accountability works
Today, decisions about Na'at are made by Sintérgica Labs and documented. A community council and an independent evaluation panel are planned, but have not been formed yet.
- Public document
- CC0 1.0
- Version 1.0 · March 2026
- 1
- Accountable today
- Sintérgica Labs
- 2
- Planned bodies
- Council and external panel, not formed yet
- 4 h
- Critical incident acknowledgment
- Anyone can report
- CC0 1.0
- License of this document
- Creative Commons, no restrictions
Who is accountable today and which bodies come next
The lab is accountable for the model. Bodies with outside participation will be created later, and this page shows the status of each one.
Central Coordination
Sintérgica Labs team
Makes decisions about the model: technical development, maintenance and version releases. Documents every decision.
Community Council
Intended for contributors, independent researchers and collaborating organizations.
Once formed, it will review substantive changes to the Constitution and governance policies.
Independent Evaluation Panel
Intended for researchers unaffiliated with Sintérgica AI, with experience in AI, ethics, linguistics or law.
Once formed, it will evaluate the model's behavior and WEIRD bias, and publish its findings.
Open participation
Any person or organization.
Incident reports, data, evaluations, documentation and improvement proposals.
Who decides what today?
Today every decision is made by the lab and documented. Once the Council exists, substantive changes will go through it.
| Decision type | Who decides | |
|---|---|---|
| Fixes with no impact on values | Central Coordination | Internal review |
| New capabilities or features | Central Coordination | Internal review and release notes |
| Changes to training data | Central Coordination | Internal review and documentation |
| Changes to the Model Constitution | Central Coordination | Documented change. Will go to the Council once it exists |
| Changes to this Governance document | Central Coordination | Documented change. Will go to the Council once it exists |
| Licensing decisions | Central Coordination | Announced with the model's release |
| Response to critical incidents | Central Coordination | Immediate action + subsequent public report |
Anyone can contribute
No affiliation with Sintérgica AI or prior approval is required. What is required is adherence to the Code of Conduct and alignment with the Model Constitution.
Data and corpus
Mexican Spanish datasets, indigenous languages, regulations, case law.
GitHub + donation form
Evaluations and benchmarks
WEIRD bias tests, performance evaluations in Mexican contexts, red teaming.
GitHub, Issues or evaluations repository
Documentation
Translations, usage guides, application cases, corrections.
GitHub, Pull Request
Incident reporting
Inappropriate behaviors, detected biases, security failures.
Dedicated reporting channel (Section V)
MAJOR.MINOR.PATCH scheme
Na'at uses adapted semantic versioning. Each component indicates the level of change and the required approval process.
2.x.x
Substantive change to the base architecture, parameter scale or fundamental values.
x.2.x
New fine-tuning cycle with additional data, new documented capabilities, expansion to new languages.
x.x.3
Behavior corrections, alignment adjustments, security fixes with no capability changes.
Incident reporting and management
An incident is any model behavior that causes harm, violates the Constitution, represents a security risk, or exhibits systematic discriminatory bias.
Primary channel
clemente.hernandez@sintergica.ai
Subject: "INCIDENT, Na'at"
Technical channel
GitHub Issues labeled "incident" or "security"
Anyone can report. Anonymous reports are accepted.
Immediate risk of serious harm, violation of absolute restrictions, exploitable security flaw.
Systematic inappropriate behavior, documented discriminatory bias, confirmed malicious use.
Behaviors violating the Constitution without immediate harm, systematically culturally incorrect responses.
Inaccuracies, suboptimal behaviors, improvement suggestions.
An independent evaluation, still to be formed
The Panel does not exist yet. Once formed, its members cannot have an active contractual relationship with Sintérgica AI.
Its reports will be public and will appear on the official channels within 30 days of delivery. Central Coordination may add technical responses, but cannot modify the Panel's findings.
WEIRD bias
Systematic evaluation of the degree to which the model reproduces Western biases in responses to Mexican and Latin American contexts.
Equity across communities
Analysis of quality differences in responses to users of different origins, languages, educational levels, and regions.
Robustness against adversarial use
Red teaming to detect ways to circumvent the Constitution's restrictions.
Accuracy in Mexican domain
Evaluation of legal, regulatory, cultural, and linguistic knowledge specific to Mexico.
Agentic behavior
Evaluation of the application of caution principles when the model operates autonomously.
Terms of use
Na'at's license is announced with its release. Its use comes with clear commitments.
Attribution
Any public or commercial use must acknowledge the model's origin: Sintérgica AI / Na'at.
Not against the Constitution
Users commit to not using the model for the prohibited behaviors described in the Model Constitution.
Publish modifications
Anyone who publishes a modified version must clearly document what changed and under what terms.
Uses explicitly prohibited under any license
- Any use that violates the absolute restrictions of the Constitution
- Training systems for non-consensual mass surveillance
- Generating disinformation for political manipulation
- Use in autonomous weapons systems
- Removing or suppressing model origin attributions
Official channels
Public rules before the model is released
The governance of an AI model is as important as its technical architecture. A model with the right values but no accountability mechanisms is as risky as one without values.
- Public document
- Documented decisions
- External evaluation planned