Palantir CTO says company helped manage Anthropic and US government friction

Navigating the Anthropic Conflict

Palantir's chief technology officer, Shyam Sankar, recently addressed the company's involvement in a public clash between Anthropic and the US Department of War. As reported by The Next Web, Palantir acted as the primary conduit for Anthropic's work within the national security system. This position put the company in a difficult spot when the two entities entered a dispute over how AI should be deployed or governed.

Sankar admitted that Palantir was essentially caught in the middle of the disagreement. Despite the friction, he stated that he does not regret the partnership with Anthropic. He noted that the various parties involved in the conflict likely did not handle the situation well, but Palantir maintained its position as a provider for both Anthropic and other frontier labs.

The tension eventually led to a shift in government procurement. Sankar informed CBS in October that the Pentagon has already moved away from using Anthropic's AI. This transition marks a significant moment in how the Department of War selects and integrates large-scale AI models into its sensitive operations.

Building Model Agnostic Tools

In response to the volatility of the AI landscape, Palantir has pivoted toward creating more flexible infrastructure for its clients. Sankar explained that the company is putting significant effort into building systems that allow customers to swap one AI model for another safely. This is intended to prevent organizations from becoming locked into a single provider's ecosystem.

For commercial customers, this flexibility is becoming a core requirement. While Palantir does not use the Chinese model DeepSeek itself, it provides the framework for its customers to choose which models they want to deploy. This approach allows businesses to experiment with different capabilities while maintaining a consistent operational backend.

The goal is to ensure that a company's entire workflow isn't disrupted if a specific model becomes unavailable or underperforms. By creating a layer of abstraction, Palantir helps its clients manage the risks associated with the rapidly changing frontier of artificial intelligence. This strategy addresses the practical needs of businesses that require reliability over brand loyalty.

Views on AI Safety and Religion

Sankar also shared his perspective on the philosophical underpinnings of the AI safety movement. He suggested that the concept of effective altruism has transformed AI safety into something resembling a secular religion. He argued that the common narrative of AI either creating a utopia or causing the end of humanity rejects the importance of human agency.

He expressed concern that these types of beliefs are often presented to the general public as technical critiques. While he stated he does not mind what individuals believe personally, he believes it is important to distinguish between faith-based views and actual engineering. He emphasized that the primary duty of developers is to engineer these systems carefully and responsibly.

This distinction is important for the future of AI governance. If safety concerns are viewed primarily through a religious or ideological lens, it may complicate the technical work of making AI reliable. Sankar's comments suggest a preference for a more grounded, engineering-focused approach to the risks posed by frontier models.

The Competitive Landscape

When discussing the global competition in AI, Sankar noted that Chinese labs are currently behind the frontier. However, he predicted a shift in their behavior as they progress. He expects that Chinese firms will eventually stop releasing open-weight models once they feel they have reached the frontier of what is technologically possible.

This prediction comes at a time when models like DeepSeek have been making headlines for their accessibility. Sankar's view suggests a looming transition where the most powerful models become increasingly closed-source. This would mirror the trends seen in Western labs, where the most advanced capabilities are often kept behind private APIs.

Finally, Sankar observed that most customers who use AI primarily to cut staff and costs have failed to achieve those goals. Instead, he noted that the customers who are actually succeeding are those using AI to win their specific markets. This suggests that the most successful applications of AI are those that focus on competitive advantage rather than simple cost-cutting.