Palantir CTO Shyam Sankar tells Stratechery he does not regret working with Anthropic despite the high-stakes friction with the US defense sector.

The intersection of artificial intelligence, national security, and corporate philosophy has rarely been as fraught as it is in the current climate. In a candid, wide-ranging interview published this Thursday with Ben Thompson for Stratechery, Palantir Technologies Chief Technology Officer Shyam Sankar addressed the company’s complex entanglement with Anthropic during its highly publicized friction with the United States Department of Defense. The discussion underscored the precarious position of tech integrators who act as the bridge between cutting-edge foundational model providers and the rigid, high-stakes requirements of government agencies.

A Strategic Collision: The Anthropic-Pentagon Nexus

For months, the relationship between Anthropic—the San Francisco-based AI research lab known for its "Constitutional AI" approach—and the U.S. national security apparatus was the subject of intense scrutiny. As an integrator, Palantir served as the conduit for deploying Anthropic’s large language models within sensitive government environments. This partnership placed Palantir squarely in the line of fire when concerns regarding safety protocols, institutional risk, and ideological misalignment between the lab and the Pentagon began to surface.

When asked by Thompson if he harbored regrets regarding the collaboration, Sankar was unequivocal. "Yes, we were stuck in the middle. I don’t regret it. We still work with Anthropic; we work with both frontier labs," Sankar stated. His response highlights a pragmatic philosophy that defines Palantir’s current market strategy: neutrality and interoperability. Rather than binding itself to the fortunes of a single model provider, Palantir has pivoted toward building a robust "model-agnostic" architecture. This allows their clients to swap out underlying AI engines as political or technical requirements shift, effectively insulating the company from the volatility of its partners.

The Chronology of the Dispute

The tension between Anthropic and the Department of Defense did not materialize overnight. The timeline of this friction began to solidify in mid-2024, as the Pentagon sought to accelerate the integration of Generative AI into intelligence and logistical workflows.

• Early 2024: Palantir begins expanding its AIP (Artificial Intelligence Platform) to include Anthropic’s Claude models, specifically tailored for government use cases in secure environments.
• Summer 2024: Reports emerge of internal friction within the Pentagon regarding the "safety-first" restrictions embedded in Anthropic’s models, which some officials argued hampered the utility of the tools in time-sensitive military operations.
• October 2024: In an interview on CBS’s Face the Nation, Sankar confirmed that the Pentagon had moved to replace Anthropic’s models with alternative solutions, signaling a shift in the procurement strategy toward models that offered more flexible operational parameters.
• November 2026: The current Stratechery interview serves as the first high-level post-mortem on the ordeal, with Sankar admitting that the situation was poorly managed by the various stakeholders involved.

The "Secular Religion" of AI Safety

One of the most provocative segments of the interview focused on the rise of "Effective Altruism" (EA) and its influence on the AI safety movement. Sankar, who had provided a list of discussion topics to Thompson in advance, explicitly questioned whether the obsession with AI existential risk has evolved into a secular religion.

"Clearly I think so," Sankar remarked. He argued that the binary framing of AI as either a path to utopian salvation or a catalyst for the end of humanity fundamentally rejects the concept of human agency. By focusing on abstract, apocalyptic scenarios, he contends that the industry neglects the concrete duty of engineers to build systems that are reliable, predictable, and useful in the present. Sankar’s concern is not the existence of these beliefs, but their weaponization. He noted that when ideological stances are presented to the public as "technical critiques," they obfuscate the practical engineering challenges that companies like Palantir face daily.

Market Realities: Efficiency vs. Transformation

Beyond the geopolitical and philosophical debates, Sankar offered a pragmatic assessment of how corporate customers are currently leveraging AI. According to his observations, there is a clear divide in the efficacy of AI adoption.

Many organizations that initially sought to utilize AI purely as a cost-cutting measure—specifically for headcount reduction—have largely failed to meet their objectives. The data suggests that AI implementations centered solely on labor arbitrage often falter due to the loss of institutional knowledge and the complexity of integrating models into legacy workflows. Conversely, Sankar noted that the clients succeeding in the current market are those utilizing AI as a strategic lever to dominate their specific industries, rather than as a simple automated accountant.

The China Factor and Open-Weight Models

The competitive landscape involving Chinese AI labs, such as DeepSeek, also featured prominently in the discussion. While Palantir maintains that it does not use Chinese-developed models internally, its platform enables commercial clients to make their own choices regarding the models they deploy.

Sankar observed that while Chinese laboratories remain behind the "frontier" of Western capabilities, the gap is narrowing. He posited a theory of strategic transparency: he expects that as soon as Chinese labs believe they have achieved true parity with Western frontier models, they will cease the release of open-weight models. This prediction is particularly timely given the recent release of DeepSeek’s "Super-Agent," which has stoked debate in Washington regarding the pace of foreign AI development and the risks of open-sourcing powerful, dual-use technologies.

Analytical Implications: The Business of Belief

The interview concluded with a pointed challenge from Thompson, who suggested that Sankar’s stance on AI safety and model neutrality is as much a function of Palantir’s business model as Anthropic’s safety-centric warnings are a function of their own market positioning.

Sankar acknowledged this alignment, noting that in the current AI era, companies are essentially "built on beliefs." Anthropic has been transparent with its investors that its strict safety protocols might make it a political target, a risk the company seemingly accepts as part of its foundational mission. For Palantir, the strategy is different: it is an attempt to remain the essential layer of the stack—the "operating system" for the modern enterprise—regardless of which specific AI model sits at the core.

Broader Impact and Conclusion

The friction between Palantir and the Pentagon regarding Anthropic serves as a microcosm for the broader tensions within the AI industry. As national security interests become increasingly inseparable from private sector innovation, the "neutrality" of the software integrator is being tested.

Sankar’s comments suggest that the industry is entering a phase of maturity where "AI safety" will no longer be a catch-all term for caution, but a subject of fierce debate regarding technical implementation versus ideological adherence. As government agencies move toward more robust, bespoke, and perhaps less "restrictive" AI models, the role of companies like Palantir will become even more critical. Their ability to navigate the shifting sands of global AI development—balancing the need for performance with the mandates of national security—will likely define the winners of the next decade of digital transformation. The "religious war" over AI, as described by Sankar, shows no signs of abating, but the frontline has shifted from abstract theory to the practical, hard-fought realities of government procurement and deployment.