A new Reddit leak surfaced an internal email branding Sam Altman as “one of the worst crooked-minded CEOs currently deciding our future,” reigniting the spotlight on OpenAI and the governance of the most talked-about company in AI. The post, by user Blue-Sea2255, shows an image of what appears to be an internal company communication, its full contents hidden. What stands out is a harsh portrait of the CEO at a time when trust in AI leadership is already strained by past scandals and sudden reversals.

The most telling detail isn’t the colorful expression used against Altman, but the fact that the criticism comes from inside the organization. It’s not the first sign of internal friction: Altman’s tumultuous ouster and swift reinstatement in November 2023 had already revealed deep board-level fractures and a climate of conflict between those pushing for commercial acceleration and those calling for caution. An email dubbing the CEO “crooked-minded” confirms those fractures have never fully healed.

For enterprises weighing how to integrate large language models (LLMs) into their workflows, such warning signs carry real weight. Relying on a cloud platform dominated by a vendor whose leader faces such harsh internal attacks raises legitimate questions about long-term stability, decision-making transparency and data protection. If OpenAI’s internal culture is so polarized, how reliable are its security policies or privacy commitments? And, crucially, what happens to models and services if another governance crisis hits?

In this context, on-premise deployment – or at least self-hosting – takes on a value beyond simple total cost of ownership (TCO) reduction. Running LLMs on your own infrastructure removes the operational fate of your AI applications from someone else’s turbulence: data remains confined, model versions aren’t subject to a vendor’s unilateral decisions, and the entire stack becomes auditable. This is not about demonizing OpenAI, which remains a technological benchmark, but about recognizing that sovereignty is the only lever that can turn an over-dependence risk into a control advantage. Those moving toward local hosting are buying insurance against the upheavals that periodically rock the most media-exposed companies.

Of course, bringing models with hundreds of billions of parameters in-house is no trivial task: it demands specialized hardware investments, orchestration skills and continuous update policies. Yet the leak targeting Altman is no mere water-cooler gossip; it’s a reminder that the governance of an AI ecosystem cannot rest solely on the good faith of a single CEO, however visionary. For anyone already weighing the trade-offs between cloud and on-premise, episodes like this add one more argument to the balance.