AI Governance

AI Agent Protocol

How they operate, what they can do, and who oversees our artificial intelligence agents.

Last updated: February 2026

Agent classification

01

Level III: High Autonomy

Ariel, Galeno

Execute multi-step tasks independently. Humans review outputs, not each intermediate step. They operate in software engineering, organizational analysis, and report generation.

Cannot:

  • × Cannot access client data without explicit authorization
  • × Cannot send external communications (emails, messages) without approval
  • × Cannot modify production databases without prior review
02

Level II: Medium Autonomy

Minerva, Socrates

Operate within defined boundaries. Generate, organize, and structure content and information. All deliverables require human approval before external distribution.

Cannot:

  • × Cannot publish content autonomously
  • × Cannot make decisions that directly affect clients
  • × Escalate to the human team when ambiguity is detected
03

Support: Low Autonomy

Eco

Activated when primary agents are unavailable. Processes basic analysis tasks with reduced action margins. All outputs require explicit human validation.

Cannot:

  • × Cannot execute actions affecting external systems
  • × Cannot generate strategic recommendations
  • × Operates exclusively as a temporary backup

Human oversight framework

04

Every decision that affects people goes through people

The human-in-the-loop principle applies to: client data, strategic recommendations, financial information, and personnel decisions. Human oversight scales proportionally to the agent's autonomy level.

05

Testing before production

Every agent capability is tested in a staging environment before being activated in production. Validations include edge cases, behavior with incomplete data, and error responses. Aligned with Anthropic's Responsible Scaling Policy.

06

Explicit action boundaries

Each agent has defined action boundaries according to its level. When an agent reaches its mandate limits, it escalates to the human team. Aligned with Anthropic's principle of protection against unintended autonomous actions.

Data protection — Law 81 of 2019

Our AI agents operate in compliance with Law No. 81 of March 26, 2019 of the Republic of Panama, which regulates personal data protection. The regulatory authority is the National Authority for Transparency and Access to Information (ANTAI).

  • Consent: Explicit consent before AI processes personal data.
  • Purpose limitation: Data used only for stated consulting purposes.
  • Data minimization: Agents access only what is strictly necessary.
  • Security: Encrypted data transmission and storage.
  • No automated decisions with legal effect: No automated decision with legal effect is made without human review (Art. 20).

Data subject rights: access, rectification, deletion, and data portability. To exercise these rights, contact info@corporatelawyersgroup.com. See our complete Privacy Policy.

Our clients' rights

  • 1 Know when AI agents participate in their project
  • 2 Request human-only work on any deliverable
  • 3 Access, correct, or delete data processed by AI agents
  • 4 Receive an explanation of how AI-generated outputs were produced
  • 5 Escalate concerns to human leadership at any time

Questions about our use of AI?

We are committed to transparency. If you want to learn more about how our agents operate, let's talk.

Schedule a conversation