Two major AI policy documents released ahead of Anthropic's latest model announcement are spotlighting a broader split over how advanced systems should be governed, deployed, and constrained.
One is a White House national security memorandum that lays out how the federal government wants to adopt and control AI in intelligence and military settings. The other is OpenAI's new public plan for ensuring artificial intelligence benefits everyone, which pairs a call for broad distribution of AI with support for tighter international coordination on frontier development.
The timing puts policy back in focus just as Anthropic unveiled Claude Fable 5, the newest version of its Claude line that the company says is suitable for wider release. Analysts and commentators are expected to spend more time on the model itself in the coming days, but the surrounding policy debate is already drawing significant attention.
The White House memorandum, National Security Presidential Memorandum-11, frames AI as a capability the administration wants to accelerate across national security missions. It sets out four pillars: adoption, adaptation, assurance, and accountability.
The first two pillars are broadly straightforward, calling for greater use of AI tools and for the government to adjust to them. The accountability section stresses that use must remain lawful and consistent with the Constitution, with responsibility placed on government officials.
The most consequential language appears under assurance. The memo says AI systems used by the national security enterprise should be reliable, steerable, and secure. It also directs agencies to ensure that commercial vendors or adversaries cannot prevent use of, disable, degrade, or materially alter systems that federal personnel rely on, unless the government knows and approves.
That language suggests the administration wants stronger control over models once they are embedded in sensitive government operations. The memo also includes a mechanism for waivers, which could allow agencies to bypass some restrictions for up to a year when national security needs require it.
In another section, the memo instructs agencies to terminate contracts with companies that repeatedly act in ways that conflict with the policy, including those firms' subcontractor arrangements. The document also asks for updates to Pentagon doctrine and calls for better testing, verification, and support for secure AI procurement.
OpenAI's plan takes a different tone. It argues that AI should be built to help people broadly, while warning that some forms of unchecked automation could be harmful. The company says AI should assist human judgment rather than replace it, and it emphasizes that people should remain central to deciding what matters.
At the same time, the plan explicitly says AI systems will increasingly assist in AI research itself. OpenAI presents that as a key factor in how fast progress will move, while acknowledging that alignment and safety research remain difficult.
The document also goes further than many industry statements in calling for international coordination. It says leading AI efforts should eventually be organized in a way that helps reduce catastrophic risk and supports the possibility of slowing frontier development when needed for safety. That position echoes similar calls previously associated with Anthropic and Google DeepMind.
The rest of OpenAI's proposal focuses on concentration of power, arguing that AI should not be controlled by a narrow group. It frames the main danger less as total loss of human control and more as the risk that the wrong people could wield very powerful systems.
Taken together, the documents show two competing impulses shaping AI policy. One is a push to expand government and commercial use of advanced systems while tightening control over how they can be modified or withheld. The other is a call to spread the benefits of AI more widely while also limiting the pace of frontier development through coordination.
Anthropic's latest release adds another layer to that debate, since the company has now placed a new model into wider circulation just as policymakers and competitors are arguing over how much freedom these systems should have, and who should be allowed to steer them.