V-PROOF Expands: Who's in Charge in a World with Millions of AI Models?
Expansión features the perspective of Gil Blancafort, co-founder of V-PROOF, on the proliferation of specialized artificial intelligence models. His contribution places one question at the center of the debate: how to demonstrate that organizations maintain control over the AI they use.
The article “Will There Be an Industry with a Million AI Models?” explores a future in which large, general-purpose models coexist with systems tailored to specific sectors, companies, and tasks.
The outlooks are not identical. Enrique Lizaso of Multiverse Computing considers that figure plausible if compressed and specialized versions are included. Juan Morán of TuringDream envisions a scenario with fewer base models and greater customization for each organization.
For V-PROOF, the question of liability remains relevant in both cases.
"Trust cannot be declared; it must be demonstrated."
Gil Blancafort, co-founder of V-PROOF, in *Expansión*.
So many agents.
Who is accountable for what they do?
Illustrative examples. The flow represents their activity and evidence; it does not represent live connections or automatic integrations.
The integrity of evidence alone does not prove that a decision is correct. Review and human judgment remain essential.
More models, more decisions to oversee
Specialization can help tailor artificial intelligence to specific processes. However, a model designed for a particular task does not eliminate the need to set limits, assign responsibility, and review its results.
At V-PROOF, we understand that governance must go hand in hand with the day-to-day use of these systems. It is not enough to know which tools an organization uses; it is also important to know which versions are active, what they are used for, and what happens when a vendor, data source, or workflow changes.
The practical question is simple: if someone reviews an AI-assisted action tomorrow, will they be able to reconstruct what happened?
Different Models, a Common Governance Framework
The report also discusses opportunities for specialization in fields such as banking, insurance, law, and healthcare, as well as the combination of generalist models and systems optimized for specific tasks.
Our take is that this combination requires oversight of the entire process. When an application uses multiple models, it is not enough to analyze each component separately. It is also necessary to understand how they are connected, what information they exchange, and who validates the result.
Let's consider a document workflow: one system extracts information, another prepares a proposal, and a person reviews it. To reconstruct that process, it is important to preserve the relationship between the inputs, the versions used, the content generated, and human approval.
Efficiency and traceability address different issues. An organization needs to address both.
From Policy to Evidence
V-PROOF's approach to verifiable governance is structured around three tasks:
Identify: understand the systems used, their purpose, and the people responsible for them.
Monitor: Establish controls and document reviews, approvals, and exceptions.
Preserve evidence: Link the activity to records that allow its integrity to be verified and the process to be reconstructed.
These tasks turn a general statement of responsibility into practices that can be reviewed.
It is also important to recognize the limitations of each piece of evidence. An integrity test can help detect modifications to a record, but it does not, on its own, prove that a response is correct, that a decision is fair, or that all applicable obligations have been met.
Technical evidence should complement human judgment and assessment, not replace them.
The V-PROOF Perspective
V-PROOF develops verifiable governance infrastructure for artificial intelligence. Our approach links AI activity, controls, and human oversight with evidence to support subsequent review.
It's not just a matter of choosing between large and small models. It's about maintaining the ability to understand, monitor, and demonstrate how each system is used within an organization.
Gil Blancafort's contribution to the debate in *Expansión* reflects that priority: beyond the number of models available, what matters is who is responsible for their use.
For us, the growth of AI must be accompanied by an equally concrete responsibility: knowing who makes the decisions, who is in control, and how that can be verified.
This post highlights V-PROOF's participation in the report and outlines our own perspective on verifiable AI governance.
The conversation continues in *Expansión*.
Your company already uses AI.
Can you demonstrate how you manage it?
Identify which systems you use, who oversees them, and what records you need to keep.
