If a digital twin is the high-fidelity “ghost” of a physical system, the underlying code is the skeleton that determines its structural integrity. As we move from the high-level architecture of these digital replicas to the accessibility of their blueprints, we find ourselves at a critical crossroads. In our previous discussions on human agency and the ethics of AI agents, we argued that open-source models are the bedrock of a transparent future. Today, that transparency is being reframed by some as a liability.
At Ambiente Ingegneria, we view this shift with a critical eye. The industry narrative is pivoting toward a “security through obscurity” model, but as engineers, we know that a skeleton with hidden fractures eventually collapses.
Is open-source AI becoming “too dangerous” for the public?
Recently, industry leaders have compared the evolution of AIโspecifically models like Claude Mythosโto the sharpening of a blade. They suggest that while earlier versions were “pocket knives,” new iterations are “machetes” capable of significant harm if left unregulated. Some have even revived the 25-year-old rhetoric once used against Linux, labeling open-source models as inherently dangerous.
We respectfully disagree. In our daily workโwhether developing integrated Machine Learning solutions or fine-tuning LLM assistantsโwe find that security is not born from “black boxes.” It is born from the same principle that governs the metric system: universal, transparent standards that everyone can verify. Hiding the logic of a model does not remove its flaws; it only removes the communityโs ability to audit and fix them.
Will transparency regulations, like the EU AI Act, stifle innovation?
There is a growing concern that while the West implements strict transparency rules (such as Article 50 of the AI Act), other global powers will continue to develop AI without such “brakes.” However, for a studio like ours, regulation is not a hurdleโit is a framework for quality.
Our commitment to data analysis and the fight against misinformation aligns perfectly with these new rules. By requiring clear documentation and origin tracking, we move away from “trust me” engineering and toward a system where the metric of success is reliability. It is much easier to spot biased data or “fake news” when you can inspect the plumbing of the model.
Is “safety” being used as a pretext for technological gatekeeping?
Yann LeCun recently warned that blocking AI development under the guise of safety is a form of “medieval obscurantism.” We agree that true technological sovereigntyโespecially for European hospitals, infrastructure, and businessesโrequires the ability to host and control our own services.
When we build web applications using Python and React, or deploy custom Odoo modules, the ability to inspect the underlying database and logic is what ensures the system is robust. True safety comes from the engineering standard of “verify, then trust,” which is only possible when the source remains open to those who build the future.