The AI Deployment Audit Playbook: An Operational, Lifecycle-Oriented Framework for the Global South

This is a practical and highly relevant governance framework for emerging regulatory environments such as Mexico. The project presents a clear structure of phases, deliverables, and responsibilities, making it easy to understand and potentially implement. Moving the framework into a real-world pilot within a public or private organization would be an excellent next step to validate assumptions, refine metrics, and demonstrate operational feasibility.

This is a good paper which aims to respond to two important challenge that exists today in AI Governance: the existence of various international non-mandatory standards and recommendations which intersect with a persisting AI regulatory gap in Mexico; and the lack of clear, actionable frameworks to implement existing standards.

The paper presents a framework which builds on key international guidelines and translates them into a operational audit phases, resulting in a tool that could be useful for AI deployers in the government and private sectors. I particularly congratulate you for incorporating sociotechnical impact assessments that include marginalized populations. The paper correctly identifies the risks of audit washing and suggest some mitigation actions.

At this stage, I would consider it a good basis that the authors may consider building on through some of these recommendations:

1. Link your work with at least some human rights key elements, for instance review the UN Guiding Principles on Business and UNDP's Human Rights Impact of AI Assessment Toolkit. Adding human rights as an element of your audit will make it more effective and more solid even in the volatile Mexican regulatory context that you identify well.

2. In addition to rightly mentioning the situation with INAI, mention briefly that there are over 150 AI-related legislative proposals in Mexico, none of which has passed; and some of which simply replicate the EU AI Act.

3. I would suggest you explicitly mention disability and afro-descendant populations along with the others you mention; as well as the an obligation to consult with representative organizations of marginalized groups. In addition to being mandatory in Human Rights Law, it is an important added value of a framework from a Global South perspective.

If you have not done so, I suggest that as you move forward you already plan with whom you need to create partnerships and/or in what forums you can test and present your playbook to start translating it into action. How would you promote adoption if it is not legally binding?

Thank you for your important work, I look forward to hearing more about how you take this forward, it is much important work in Mexico and beyond.

It says no instrument offers a step-by-step audit workflow, yet only compares against the standards it integrates, not the auditing literature that already describes the same phase-ordered audit.

Three things weaken it further. There is no evidence it works, since the two walkthroughs were built by the same people who designed the workflow and cannot fail by construction. The math is not useful, since the approval formula just means "both scores must pass" with placeholder thresholds and no way to compute them. And the Global South framing is thin, since the playbook is built almost entirely from Northern instruments and the only regional content is Phase 6.

The team is candid about most of this, and chaining scattered standards into one executable sequence has practical value. I would reframe the work as operationalization rather than novelty, situate it against the auditing literature, let the Mexican context shape the method rather than just motivate it, and at minimum test it against a case the team did not design itself, even a desk audit of a real documented system, to break the circularity of the current walkthroughs.

The paper is very clearly written. As it's an adaptation and organization of existing instruments for the Mexico context, the contribution is incremental but useful. The justification for why the proposed framework would be especially adequate for volatile regulatory contexts appears insufficient, which is a concern given that this is a central claim of the study. The examples in the results section are instructive but could have been more detailed.