Sacks and Chamath Support Karp's Warning: Companies Should Not Share Proprietary Data with OpenAI/Anthropic
On the All-In Podcast, David Sacks and Chamath Palihapitiya agreed with Palantir CEO Alex Karp's warning that companies face significant risks when transmitting trade secrets, know-how, and customer data to AI providers like OpenAI and Anthropic, as these companies may later use the data to compete.
They cited Anthropic's launch of Claude Design (which caught partner Figma off guard) and Claude Code as examples of vertical applications, emphasizing that companies should maintain control over their data, model weights, and computing power to avoid being trapped in a "walled garden."
This discussion highlights the importance of data sovereignty in the era of large models.
Source: Public Information
ABAB AI Insight
Alex Karp has previously criticized the data practices of closed-source AI companies. The support from Sacks and Chamath, as prominent investors and entrepreneurs, reinforces the business community's concern over data leakage and competitive risks.
In terms of capital pathways, companies are shifting towards self-owned AI infrastructure or open-source solutions to retain data advantages. Companies like Palantir that emphasize data control may benefit from this shift, strategically pushing enterprise AI from external reliance to internal construction.
Similar warnings have been issued by Mistral founder Arthur Mensch, indicating a trend towards companies building private LLMs. This incident reflects that data has become a core competitive resource in the AI supply chain, intensifying conflicts of interest between closed-source vendors and corporate clients.
Essentially, this is a game of data capital. AI training relies heavily on high-quality corporate data, leading capital to concentrate on companies that possess data sovereignty and computational control, creating new competitive barriers.
ABAB News · Cognitive Laws
Once data leaks, it becomes a weapon for others in competition.
Corporate data sovereignty is more important than the convenience of model invocation.
The explosion of AI vertical applications amplifies the risk of data leakage.