Tech workers urge DOD, Congress to withdraw Anthropic label as a supply chain risk

Hundreds of tech workers person signed an open letter urging the Department of Defense to retreat its designation of Anthropic arsenic a “supply concatenation risk.” The missive besides calls connected Congress to measurement successful and “examine whether the usage of these bonzer authorities against an American exertion institution is appropriate.”

The missive includes signatories from large exertion and task superior firms including OpenAI, Slack, IBM, Cursor, Salesforce Ventures, and more. It follows a dispute betwixt the DOD and Anthropic aft the AI laboratory past week refused to springiness the military unrestricted entree to its AI systems. 

Anthropic’s 2 reddish lines successful its negotiations with the Pentagon were that it didn’t privation its exertion to beryllium utilized for wide surveillance connected Americans oregon to powerfulness autonomous weapons that made targeting and firing decisions without a quality successful the loop. The DOD said it had nary plans to bash either of those things, but that it didn’t judge it should beryllium constricted by the rules of a vendor. 

In effect to Anthropic CEO Dario Amodei’s refusal to cave to Hegseth’s threats, President Donald Trump connected Friday directed national agencies to halt utilizing Anthropic’s exertion aft a six-month modulation period. Hegseth said helium would marque bully connected his threats and designate Anthropic a proviso concatenation hazard — a designation usually reserved for overseas adversaries that would blacklist the AI steadfast from moving with immoderate bureau oregon institution that does concern with the Pentagon. 

In a post connected Friday, Hegseth wrote: “Effective immediately, nary contractor, supplier, oregon spouse that does concern with the United States subject whitethorn behaviour immoderate commercialized enactment with Anthropic.” 

But a station connected X does not automatically marque Anthropic a proviso concatenation risk. The authorities needs to implicit a hazard appraisal and notify Congress earlier subject partners person to chopped ties with Anthropic oregon its products. Anthropic said successful a blog post the destination is some “legally unsound” and that it would “challenge immoderate proviso concatenation hazard designation successful court.”

Many successful the manufacture spot the administration’s attraction of Anthropic arsenic harsh and wide retaliation. 

“When 2 parties cannot hold connected terms, the mean people is to portion ways and enactment with a competitor,” the unfastened missive reads. “This concern sets a unsafe precedent. Punishing an American institution for declining to judge changes to a declaration sends a wide connection to each exertion institution successful America: judge immoderate presumption the authorities demands, oregon look retaliation.” 

Beyond interest implicit the government’s harsh attraction of Anthropic, galore successful the manufacture are inactive acrophobic astir imaginable authorities overreach and usage of AI for nefarious purposes. 

Boaz Barak, an OpenAI researcher, wrote successful a societal media post connected Monday that blocking governments from utilizing AI to bash wide surveillance is besides his “personal reddish line” and “it should beryllium each of ours.”

Moments aft Trump publically attacked Anthropic, OpenAI announced it had reached a deal of its ain for its models to beryllium deployed successful the DOD’s classified environments. OpenAI CEO Sam Altman said past week that the steadfast has the aforesaid reddish lines arsenic Anthropic.

“If thing bully tin travel retired of the events of the past week, it would beryllium if we successful the AI manufacture commencement treating the contented of utilizing AI for authorities maltreatment and surveilling its ain radical arsenic a catastrophic hazard of its ain right,” Barak wrote. “We person done a bully occupation of evaluations, mitigations, and processes, for risks specified arsenic bioweapons and cyber security. Let’s usage akin processes here.”