AI TRENDS | David Sacks Says Enterprise AI Security Depends on Control of Data, Models, and Compute

David Sacks said enterprise AI security is primarily about companies maintaining full control over their own data, model weights, and computing infrastructure, rather than relying on abstract AI alignment research or government-style regulatory frameworks. According to Odaily, Sacks made the comments on X while discussing an interview involving Palantir CEO Alex Karp.

Sacks said some traditional media characterized Karp’s remarks as “emotional,” but he argued the interview raised core issues in enterprise AI security. He said companies need control to prevent their key knowledge assets from being absorbed by model vendors and converted into the vendors’ product advantages.

Citing Karp’s view, Sacks said enterprise customers focus on ownership and control of computing resources, models, and the data stack, describing it as ensuring that “ownership of the means of production is not transferred.”

Sacks pointed to a reported dispute involving Figma’s cooperation with Anthropic. He said media reports claimed Anthropic caught its partner off guard when it launched Claude Design and was accused of expanding into the application-layer domain where ecosystem partners operate, changing how value is captured.

He added that similar patterns were seen in product expansions such as Claude Code and Claude Legal, where model capabilities extend upward into vertical application areas.

Sacks said this trend suggests model vendors are shifting from being base-model providers to becoming competitors in vertical applications, increasing vendor-dependence risks for enterprise customers. He said enterprise AI security is not about trusting long-term commitments from model vendors, but about ensuring choice and control at the model layer to protect company data and business “alpha.”