New rules for US national security agencies balance AI's promise with need to protect against risks

WASHINGTON -- WASHINGTON (AP) — New rules from the White House connected nan usage of artificial intelligence by U.S. nationalist information and spy agencies purpose to equilibrium nan technology's immense committedness pinch nan request to protect against its risks.

The model signed by President Joe Biden and announced Thursday is designed to guarantee that nationalist information agencies tin entree nan latest and astir powerful AI while besides mitigating its misuse.

Recent advances successful artificial intelligence person been hailed arsenic perchance transformative for a agelong database of industries and sectors, including military, nationalist information and intelligence. But location are risks to nan technology's usage by government, including possibilities it could beryllium harnessed for wide surveillance, cyberattacks aliases moreover lethal autonomous devices.

“This is our nation’s first-ever strategy for harnessing nan powerfulness and managing nan risks of AI to beforehand our nationalist security,” nationalist information advisor Jake Sullivan said arsenic he described nan caller argumentation to students during an quality astatine nan National Defense University successful Washington.

The model directs nationalist information agencies to grow their usage of nan astir precocious AI systems while besides prohibiting definite uses, specified arsenic applications that would break constitutionally protected civilian authorities aliases immoderate strategy that would automate nan deployment of atomic weapons.

Other provisions promote AI investigation and telephone for improved information of nan nation's machine spot proviso chain. The rules besides nonstop intelligence agencies to prioritize activity to protect nan American manufacture from overseas espionage campaigns.

Civil authorities groups person intimately watched nan government's expanding usage of AI and expressed interest that nan exertion could easy beryllium abused.

The American Civil Liberties Union said Thursday nan authorities was giving excessively overmuch discretion to nationalist information agencies, which would beryllium allowed to “police themselves.”

“Despite acknowledging nan sizeable risks of AI, this argumentation does not spell astir acold capable to protect america from vulnerable and unaccountable AI systems," Patrick Toomey, lawman head of ACLU’s National Security Project, said successful a statement. “If processing nationalist information AI systems is an urgent privilege for nan country, past adopting captious authorities and privateness safeguards is conscionable arsenic urgent."

The guidelines were created pursuing an ambitious executive bid signed by Biden past twelvemonth that called connected national agencies to create policies for really AI could beryllium used.

Officials said nan rules are needed not only to guarantee that AI is utilized responsibly but besides to promote nan improvement of caller AI systems and spot that nan U.S. keeps up pinch China and different rivals besides moving to harness nan technology's power.

Sullivan said AI is different from past innovations that were mostly developed by nan government: abstraction exploration, nan net and atomic weapons and technology. Instead, nan improvement of AI systems has been led by nan backstage sector.

Now, he said, it is “poised to toggle shape our nationalist information landscape.”

Several AI manufacture figures contacted by The Associated Press praised nan caller policy, calling it an basal measurement successful ensuring America does not output a competitory separator to different nations.

Chris Hatter, main accusation information serviceman astatine Qwiet.ai, a tech institution that uses AI to scan for weaknesses successful machine code, said he thought nan argumentation should pull bipartisan support.

Without a argumentation successful place, nan U.S. mightiness autumn down connected nan “most consequential exertion displacement of our time.”

“The imaginable is massive,” Hatter said. “In subject operations, we’ll spot autonomous weaponry — for illustration nan AI-powered F-16 and drones — and determination support systems augmenting quality intelligence."

AI is already reshaping really nationalist information agencies negociate logistics and planning, amended cyber defenses and analyse intelligence, Sullivan said. Other applications whitethorn look arsenic nan exertion develops, he said.

Lethal autonomous drones, which are tin of taking retired an force astatine their ain discretion, stay a cardinal interest astir nan subject usage of AI. Last year, nan U.S. issued a declaration calling for world practice connected mounting standards for autonomous drones.