Artificial Intelligence as Powerful as a 'Nuclear Bomb': A New Weapon Confusing the Pentagon

The United States has entered a new and confusing phase in the AI arms race, after advanced AI models evolved from helper tools into potential capabilities for penetrating enemy networks and testing military defense systems, amid growing American fears that China could possess similar capabilities within months.

According to The New York Times, this shift has caused confusion within U.S. national security institutions, clearly evident in the clash between the Pentagon and Anthropic, as the Department of Defense’s desire to remove the company’s technologies from its systems clashed with the need for an advanced AI model that could become a decisive weapon in cyber warfare.

Contradictory Orders

According to the newspaper, senior U.S. military contractors received instructions from the Air Force mid-July, demanding they eliminate all use of Anthropic products in weapons software and control systems by September 1, under threat of jeopardizing their work with the Pentagon.

But about a month later, a reverse message arrived, asking them to pause and refrain from removing the company’s products for now, reflecting the confusion gripping Washington’s response to the accelerating pace of AI development.

The crisis stems from an open dispute between Defense Secretary Pete Hegseth and Anthropic’s CEO Dario Amodei over restrictions on the military use of the company’s technologies.

Amodei demanded guarantees that his company’s products would not be used for domestic surveillance of Americans or for creating fully autonomous weapons without human oversight, but the Pentagon rejected these conditions and labeled the company a "threat to the supply chain," before banning its products for military use.

Unilateral Disarmament

The irony emerged when Anthropic unveiled its "Mythos" model, designed to detect and exploit software vulnerabilities, placing the military institution in an unexpected dilemma: Washington had banned itself from using a technology that could become one of the most powerful new cyber weapons.

The newspaper quoted a former U.S. official saying that abandoning Mythos would amount to "unilateral disarmament," given its potential in offensive cyber operations.

Despite the Pentagon’s ban on using "Claude" for intelligence analysis and code writing, the National Security Agency (NSA) was granted an exception to use experimental versions of Mythos to test U.S. military network defenses and identify potential weaknesses in adversary networks, including those of China, Russia, North Korea, and Iran.

China Within Months

The concern extends beyond the conflict with Anthropic, as American experts believe China may lag behind the United States by only around six months—or less—in developing the most advanced AI models, despite maintaining a gap in manufacturing advanced semiconductors.

Washington fears these capabilities could give Beijing more powerful tools to breach U.S. communication, water, and power networks, especially following attacks like "Volt Typhoon" and "Salt Typhoon," while former officials warn that AI is now capable of amplifying cyberattack capabilities in unprecedented ways.

The level of concern has reached the point where CIA Director John Ratcliffe compared AI capabilities to "digital nuclear weapons," as U.S. institutions have spent billions of dollars integrating this technology into weapons development, military network testing, and ensuring that weapon-related systems, including those tied to nuclear weapons, can withstand AI-driven hacking attempts.