Access the Editor’s Summary without charge
Roula Khalaf, the FT’s Editor, curates her preferred articles for this recurring publication.
To bolster its operational capacity in any prospective military confrontation with Beijing, the Pentagon aims to develop AI-driven cybernetic instruments for pinpointing China’s critical infrastructure assets.
Multiple sources privy to the plans indicated that the agency was engaged in discussions with prominent AI firms regarding collaborations to execute autonomous surveillance of China’s electricity networks, public services, and crucial digital systems, along with those of other rivals.While the US has already fashioned formidable digital spying instruments, it now strives to utilize novel AI-enabled mechanisms to detect software vulnerabilities within adversaries’ networks. These weaknesses could then be leveraged to facilitate greater penetration and compromise those systems during any future hostilities.
Sources further mentioned that the envisioned framework would employ AI to infiltrate digital systems, chart weaknesses, and incorporate prospective objectives into US military strategy. The Pentagon, however, refused to elaborate.
Agreements valued at approximately $200 million have been granted to OpenAI, Anthropic, Google, and Elon Musk’s xAI, enabling them to collaborate with the US administration on defense, digital, and protection applications. The specific firms that will participate in this novel digital undertaking remain undecided.
Following CEO Dario Amodei’s statement that he would decline a “final offer” regarding the conditions for cooperating with the military, a high-ranking US official cautioned on Thursday that the administration would annul all its current contracts with Anthropic should the company fail to secure an arrangement with the Pentagon.
This Pentagon endeavor underscores Washington’s acknowledgment of the growing significance of digital warfare in any future conflict with China, along with the belief that AI could shift the advantage. However, this development also emerges amidst elevated friction with some of the nation’s cutting-edge AI companies concerning the permissible extent of their technology’s application in military operations.
Dennis Wilder, the ex-chief of China intelligence at the CIA and now at Georgetown University, asserted that AI digital instruments would assist in resolving the challenge presented by the extensive volume of human resources required to scrutinize and pinpoint susceptible infrastructure.
“It’s comparable to a nocturnal burglar who tests the front entrances of homes until discovering one that has been left unsecured,” Wilder, now at Georgetown University, commented. “AI-supported digital intrusion can dramatically multiply the quantity of entry points checked, thereby enabling significantly more effective and precise charting of objectives for targeting.”
While defense digital specialists currently perform duties pinpointing susceptible objectives, the prospective AI mechanism would execute these operations more rapidly and at a greater scale, with significantly reduced human input.
“They have been developing digital assault tactics for penetrating electricity networks. It is essential to craft both offensive and defensive capabilities; one cannot exist without the other,” stated one individual knowledgeable about the initiatives.
A different source mentioned that energy facilities situated close to data hubs might be selected to impair opponents’ artificial intelligence capacities.
Secretary of Defense Pete Hegseth has pursued entry to potent generative AI technology for what he characterizes as “lawful use.” However, AI laboratories have been reluctant to grant the Pentagon unrestricted authority over this technology.
Company insiders revealed that Anthropic had aimed to prevent the deployment of its Claude system in deadly self-governing armaments and desired limitations on AI application for widespread internal monitoring,
Hegseth informed Amodei on Tuesday that if his company did not consent to his conditions by Friday, it could be labeled a procurement danger or have its technology commandeered by the government.
Multiple sources familiar with the issue indicated that while Claude is presently the sole AI model employed for secret missions, personnel at other AI research facilities, including OpenAI and Google, have voiced analogous worries.
“This serves as a reaction to what is emanating from China and the absence of safeguards there,” stated one individual knowledgeable about the Defense Department’s position. “It entails unrestricted deployment; we cannot have constraints upon us when everything commences.”