Odaily Planet Daily reports that on X, David Sacks commented on related interviews with Palantir CEO Alex Karp. He said some traditional media interpreted them as “emotional remarks,” but in fact his views reveal the core issue of enterprise AI safety.

Sacks said that real enterprise AI safety is not an abstract “AI alignment” research agenda or a government-style regulatory framework. Rather, it is the ability of enterprises to fully control their own data, model weights, and compute infrastructure—so as to prevent core knowledge assets from being absorbed by model vendors and turned into advantages for the vendors’ own products.

He cited Karp’s view, saying that what enterprise customers truly care about is control over compute resources, models, and their data stack—namely, that “ownership of the means of production is not transferred.”

Sacks also cited an example from the controversy over Figma’s cooperation with Anthropic. According to media reports, when Anthropic launched Claude Design, it “caught its partners off guard,” and was accused of moving into the application-layer areas where its ecosystem partners operate during product expansion, causing the value-capture structure to change.

He further pointed out that similar patterns also appear in expansions of product lines such as Claude Code and Claude Legal, extending model capabilities upward into vertical application domains.

Sacks believes this trend means model vendors are shifting from “foundation model providers” to “competitors in vertical applications,” exposing enterprise customers to greater vendor lock-in risk. The essence of enterprise AI safety is not trusting a model vendor’s long-term commitments, but ensuring that there is optionality and control at the model layer—thereby protecting its own data and business “alpha”.