Politics
Pentagon Approves AI Vendors That Never Once Asked If This Was A Good Idea
By Mor Michael, . BSN Network. Satire.
A federal appeals court ruled Wednesday that the Pentagon may continue treating Anthropic, the artificial intelligence company founded specifically to make AI less dangerous, as a danger.
The U.S. Court of Appeals for the D.C. Circuit sided with the Trump administration, allowing the Defense Department to maintain its designation of Anthropic as a supply chain risk. The court did not specify which part of "we should probably be careful about this" constituted the risk.
Anthropics's offense, as best reconstructed from the record, is that it publishes research about AI going wrong, employs people whose job titles include the word "safety," and once released a model that would decline to help you build a bioweapon. The Pentagon has reviewed these facts and determined that this is the behavior of a suspect vendor.
Defense Secretary Pete Hegseth, who oversees a department currently integrating AI decision-support tools into weapons targeting systems, said through a spokesperson that the administration remains "committed to AI leadership" and that leadership "requires the full confidence of our industrial partners." He did not define confidence. He did not need to.
The approved vendors, whose contracts remain intact, include several companies that have never once published a paper suggesting the technology might need guardrails. Their researchers describe the models as tools. They do not hold symposiums. They do not release alignment benchmarks. They have, in this sense, given the government no reason for concern.
Anthropics's CEO Dario Amodei told employees in an internal message obtained by multiple outlets that the company would continue its work "regardless of how institutions choose to categorize it." He added that safety research "speaks for itself." It does, apparently, speak loudly enough to get you blacklisted.
The company had argued that the Pentagon's designation was arbitrary and unsupported by the evidence. The court found that the government had followed the correct procedure for labeling a company a supply chain risk, which is a different question than whether the company is one, and the court was careful not to confuse the two.
Anthropics currently has no Defense contracts to lose. It was blacklisted before it had any, which is a timeline that should trouble more people than it has.
The ruling was welcomed by several defense technology firms, none of which responded to requests for comment, possibly because commenting on things is how this started.
The story we are making fun of: https://thehill.com/newsletters/technology/6112510-appeals-court-sides-with-trump-admin-in-anthropic-blacklisting/