Hegseth demands full military access to Anthropic's AI model, sets deadline

2 hours ago 5

Trust is breaking down betwixt the Pentagon and Anthropic implicit the usage of its AI model, sources acquainted with the concern told CBS News. 

In a gathering astatine the Pentagon Tuesday morning, Defense Secretary Pete Hegseth gave Anthropic's CEO Dario Amodei until the extremity of this week to springiness the subject a signed papers that would assistance afloat entree to its artificial quality model, the sources said. 

Officials are considering invoking the Defense Production Act to marque Anthropic adhere to what the subject is seeking, they said. Axios reported earlier connected immoderate of what transpired successful the meeting.

Defense officials privation afloat power of Anthropic's AI exertion for usage successful its subject operations, sources told CBS News. The institution was awarded a $200 cardinal declaration by the Pentagon successful July to make AI capabilities that would beforehand U.S. nationalist security.

Anthropic has repeatedly asked the Defense Department to hold to guardrails that would restrict the AI model, called Claude, from conducting wide surveillance of Americans, sources said. Defense officials noted that that's amerciable and said the subject is simply asking for a licence to usage the AI strictly for lawful activities.

Amodei besides wants to guarantee Claude is not utilized by the Pentagon for last targeting decisions successful subject operations without immoderate quality involvement, 1 root acquainted with the gathering said. Claude is not immune from hallucinations and not reliable capable to debar perchance lethal mistakes, similar unintended escalation oregon ngo nonaccomplishment without quality judgment, the idiosyncratic said. 

But a elder Pentagon authoritative said: "This has thing to bash with wide surveillance and autonomous weapons being used. The Pentagon has lone fixed retired lawful orders."

The authoritative said Grok, which is owned by Elon Musk's xAI, is connected committee with being utilized successful a classified setting, and different AI companies are close.

In Tuesday's meeting, Hegseth told Amodei that erstwhile the authorities purchases Boeing planes, the aerospace institution has nary accidental successful however the Pentagon uses the planes. He argued the aforesaid should beryllium existent for the military's usage of Claude.

After Amodei left, officials discussed whether to usage the Defense Production Act successful this situation, which enables the authorities to exert power implicit home industries. 

But due to the fact that officials accidental they aren't definite the authorities tin spot Anthropic astatine this point, the Pentagon whitethorn determine to officially designate the institution arsenic a "supply concatenation risk" to propulsion them retired of government, 2 sources said. Anthropic was the archetypal tech institution authorized to enactment connected the military's classified networks. 

An Anthropic spokesperson said successful a statement, "We continued good-faith conversations astir our usage argumentation to guarantee Anthropic tin proceed to enactment the government's nationalist information ngo successful enactment with what our models tin reliably and responsibly do."

Hegseth gave Anthropic a deadline of 5 p.m. Friday.

In:

What goes into investigating Anthropic's AI

Anthropic knows AI comes with risks. What it says it's doing to effort to mitigate them. 13:51

Anthropic knows AI comes with risks. What it says it's doing to effort to mitigate them.

(13:51)

Read Entire Article