David Sacks backs Alex Karp’s warning: enterprises should not hand proprietary data to frontier AI labs

David Sacks, co-founder of Craft Ventures and a long-time Silicon Valley investor, endorsed Palantir CEO Alex Karp’s warning that enterprises should be cautious about feeding proprietary data into frontier AI labs such as OpenAI and Anthropic. Sacks made the remarks during an episode of the All-In Podcast that aired Friday, aligning with Karp’s argument that the risk is less about theoretical “alignment” and more about practical control of a company’s information and systems.

In Sacks’s telling, the defining element of “AI safety” for businesses is ownership and control, including control over compute, models, the data stack, and what he referred to as “alpha,” meaning a firm’s proprietary knowledge and competitive advantage. He argued that these controls determine whether a company’s internal know-how remains within its own boundaries rather than being absorbed into another party’s ecosystem.

Sacks also said that when companies rely on frontier labs, those labs can incorporate clients’ technical know-how into future offerings. The concern, as he described it, is that a client’s learnings could be turned into competing products, leaving the original company exposed if it does not control its own AI infrastructure outright.

The discussion was framed by Sacks as a rejection of abstract alignment research in favor of operational realities, according to the reporting. A separate recap of his comments on X, noted in the coverage, made the same general case: that for enterprises, the key decision is whether they can keep their proprietary capabilities from being generalized by third-party AI providers.

Palantir sits at the center of a broader debate about how AI should be deployed in enterprise settings. While companies across industries are moving toward using large AI models, Karp’s warning, as echoed by Sacks, emphasizes that data-sharing decisions can shape competitive outcomes. The thrust of the argument is that the supplier of frontier models may learn from customer input, while the customer’s strategic advantage may be diluted if it does not retain control over where its data and derived insights go.

For Palantir investors, the remarks land in a familiar context: Palantir has long positioned itself around enterprise systems that keep sensitive workflows and data under customer governance, rather than outsourcing core operations to third parties. Sacks’s comments effectively translate that positioning into a governance and control message, focusing on compute, models, and the proprietary knowledge companies use to differentiate.

Still, the podcast discussion as reported does not offer new, company-specific disclosures about Palantir customer agreements, technical deployment details, or quantified impacts. The claims are presented as an argument about enterprise risk and control, not as evidence of specific data leaks or contractual disputes involving OpenAI, Anthropic, or Palantir.

What to watch next is whether the debate moves from commentary to measurable procurement behavior. If enterprises increasingly demand tighter controls around model usage and data governance, vendors that can offer enterprise-managed AI stacks could benefit. Conversely, if frontier labs continue to expand enterprise offerings with stronger contractual and technical safeguards, the outcome may depend on how those safeguards are structured and enforced. For now, the takeaway is that leading tech voices are pressing companies to treat “AI safety” as an issue of control, not only of model behavior.