There is a deadline in the unheated, sterile corridors of the Pentagon tapping over 5:01 p.m. this Friday. However, the founders of Anthropic already took their step 3,000 miles away in San Francisco. In one of the most breathtaking twists to a months-long feud, the CEO of the AI firm Anthropic, Dario Amodei, said on Thursday that the company would not yield to the ultimate ultimatum of the Department of Defense, which demanded that it remove the ethical guardrails on one of its AI, Claude, that the company is not allowed to use in any lethal autonomous weapons and in the domestic surveillance of its citizens masse.
The Red Lines: Why Anthropic is Digging In
The conflict revolves around two bright red lines which Anthropic does not want to cross. Competitors such as OpenAI and Google have reportedly shifted to more liberal all lawful use agreements with the Pentagon, but Anthropic has been recalcitrant.
The Ghost of Mass Surveillance
Anthropic states that the present-day AI can pattern match in disparate datasets with dangerously good results. Amodei issued his warning in a statement by saying that in the absence of protection, the military might stitch together a complete picture of any individual life by scraping the movement information, web surfing, and social connections all without a warrant. This is not just a technical risk to Anthropic but is, in essence, a threat to the democratic liberties.
The Human-in-the-Loop Doctrine
The second point of conflict is deadly autonomy. Although the Pentagon denies its interest in so-called killer robots, shooting without human authorization, it is not ready to sign a contract that specifically forbids their development. The engineers of Anthropic believe that the Large Language Models (LLMs) of the present-day still have a tendency to hallucinate and do not have the critical thinking ability of a trained soldier. They say it is a plan to disastrous mistake to use them to draw a trigger.
The Amodei added that even the best AI systems are not ready to be relied on to operate deadly weapons without a human being in the final position of control. An announcement on 1st June stated that it would not knowingly supply a product that endangered the lives of warfighters and civilians in America.
The Ultimatum of the Pentagon: AI Will not be Woke
Across the table is Defense Secretary Pete Hegseth, whose tenure has been based on the principles of purging ideological cuffs off the military. To the Pentagon brass, the safeguards provided by Anthropic are not safety; it is roadblocks, which will cost lives in a high speed confrontation with an enemy such as China.
The defense rhetoric has become very personal. Recently, Undersecretary Emil Michael claimed that Amodei had a god complex because a privately-held CEO was not entitled to make legal decisions on behalf of the U.S. military. Hegseth gave an ultimatum, and it was simple: Sign the “all lawful purposes” document before Friday evening or suffer the consequences of scorched-earth.
