US Military AI Push Stumbles Over Anthropic Feud as China Advances

Artificial intelligence companies are buying up vast quantities of used and rare books, scanning their contents for training data, and then discarding the originals, according to reports that have alarmed booksellers and cultural preservationists. The practice has drawn particular scrutiny when it involves irreplaceable volumes from earlier centuries, raising questions about whether the rush to build more powerful systems justifies the permanent loss of physical records.

At the same time, the U.S. military is pushing aggressively for AI dominance in national security, only to encounter internal contradictions and resistance from the very companies developing the technology. Recent directives from the Air Force instructed major contractors to purge software linked to Anthropic by early September, a requirement that was quietly reversed weeks later. Defense officials have cited the need to integrate advanced models rapidly, yet the back-and-forth has left weapons programs and data systems in limbo.

Anthropic chief executive Dario Amodei has publicly clashed with Defense Secretary Pete Hegseth over limits the company sought to place on military applications of its Mythos model. The restrictions reflect broader corporate concerns about how the technology might be deployed, even as the Pentagon competes with China to field AI-enhanced systems. Agencies still reliant on equipment designed decades ago now face pressure to adopt tools whose capabilities shift every few months.

The book-scanning efforts underscore a parallel pattern of short-term extraction. Companies argue that exposure to classic literature will improve model quality and reduce reliance on low-grade internet content. Yet the destruction of originals leaves no fallback if digital copies prove incomplete or if future needs demand the physical artifacts themselves. Collectors and historians note that many discarded titles cannot be replaced, turning a one-time data grab into an irreversible cultural cost.

Military planners have framed AI integration as essential to counter Chinese advances, but the administration’s handling of the issue has produced mixed signals. Contractors received conflicting guidance on Anthropic products, illustrating how policy disputes can delay operational readiness. The National Security Agency has highlighted the value of certain models for identifying system vulnerabilities, yet the same models remain subject to usage constraints that the Pentagon has alternately enforced and suspended.

Critics of unchecked AI expansion point to these episodes as evidence that commercial incentives and national security goals are colliding without adequate safeguards. The decision to treat rare books as disposable feedstock mirrors a wider tendency to prioritize scale over stewardship. Meanwhile, the military’s desire for dominance risks entangling civilian firms in applications they view as ethically fraught, even as rivals abroad move forward with fewer internal restraints.

Preservation advocates argue that scanning need not require disposal, and that institutions could retain originals for verification or future study. Military leaders counter that speed is paramount in an era when model performance doubles in short intervals. Both sides acknowledge the underlying tension: technologies built on humanity’s accumulated knowledge are being developed in ways that risk erasing portions of that same inheritance while fueling new forms of geopolitical competition.