发生了什么
The Pentagon's CTO publicly dismissed existential AI risks while detailing the transfer of classified AI operations away from Anthropic to other providers as the legal dispute over military usage restrictions continues in federal court.
Emil Michael, the US Department of Defence’s chief technology officer, spoke at a defence industry conference in Washington on Thursday, September 10, 2026. He explicitly rejected warnings about catastrophic AI risks, describing fears of mass unemployment and uncontrollable AI as a 'doom loop' fueled by general anxiety over rapid technological change. These remarks were made shortly after former Anthropic engineer Jacob Coxon posted a widely viewed warning about existential threats from self-improving AI.
Simultaneously, Michael provided concrete updates on the Pentagon's operational shift away from Anthropic. He stated that approximately 90% of the department's classified AI workload has already been transferred to other providers. The remaining work is scheduled to be moved by the end of September 2026. He also confirmed that work related to the Maven Smart Systems programme, associated with Palantir, was moved away from Anthropic months prior.
The operational shift is part of a broader legal and policy dispute. Negotiations between the Pentagon and Anthropic broke down earlier in 2026 after the department rejected safeguards proposed by the company, arguing they interfered with military authority. The Pentagon subsequently designated Anthropic as a supply-chain risk. Anthropic has filed lawsuits in California and Washington, DC. A California judge ruled in Anthropic's favor, but the case before the US Court of Appeals for the DC Circuit remains unresolved. Michael characterized the California filing as 'forum-shopping' and indicated the Pentagon expects a DC Circuit ruling in the coming months.
为什么这很重要
This development highlights a critical divergence in US government policy regarding and military deployment. By shifting classified workloads and challenging Anthropic's legal standing, the Department of Defence is asserting that operational flexibility outweighs developer-imposed safeguards. This sets a precedent for how the US military will integrate advanced AI, potentially prioritizing speed and capability over the restrictive conditions sought by AI developers, which could influence future regulatory frameworks and industry-government relations.
The Pentagon's decision to decouple its classified AI operations from Anthropic signals a prioritization of unrestricted military deployment over developer-imposed safety constraints. This move effectively tests the limits of corporate control over AI models used in national security contexts.
Michael's public dismissal of existential AI risks contrasts with the legal actions taken by the department, suggesting a strategic stance that views AI as a tool for operational advantage rather than a domain requiring strict ethical . This could influence other government agencies and private sector partners in their approach to AI integration.
The ongoing legal battle in the DC Circuit is likely to establish a significant precedent regarding the balance between national security needs and the contractual or ethical restrictions AI developers may attempt to impose on their technology. The outcome will have lasting implications for the AI industry's relationship with the US military.
互动机制：它实际上是如何运作的
以交互方式探索这一发展背后的基础技术。
Why can ethical evaluation not be reduced to one model score?
接下来看什么
Monitor the upcoming ruling from the US Court of Appeals for the DC Circuit, which the Pentagon expects to decide in the coming months. Additionally, track the completion of the remaining 10% of classified workload transfers by the end of September and any further statements from Commerce Secretary Howard Lutnick regarding the broader administration's stance on Anthropic.
The US Court of Appeals for the DC Circuit is expected to issue a ruling in the coming months, which will determine the legal standing of the Pentagon's designation of Anthropic as a supply-chain risk and the enforceability of the company's proposed safeguards.
The completion of the transfer of the remaining 10% of classified AI workloads by the end of September 2026 will mark the full operational separation between the Pentagon and Anthropic for these specific tasks.
Further statements from other US government officials, such as Commerce Secretary Howard Lutnick, may clarify whether the Pentagon's confrontational approach is an isolated departmental policy or part of a broader administration strategy regarding .