A company that used artificial intelligence to mimic Joe Biden’s voice and make misleading phone calls to voters in New Hampshire agreed Wednesday to pay a $1 million fine and improve its caller identification and authentication features, U.S. regulators said.

Lingo Telecom, the voice service provider that transmitted the robocalls, agreed to the settlement to resolve enforcement actions by the Federal Communications Commission, which had initially sought a $2 million fine.

Meanwhile, Steve Kramer, a political consultant who orchestrated the calls, still faces a proposed $6 million fine from the FCC and state criminal charges. The case is seen by many as a troubling early example of how AI can be used to influence groups of voters and democracy as a whole.

The phone messages were sent to thousands of voters in New Hampshire on January 21. They featured a voice similar to Biden’s, falsely suggesting that voting in the state’s presidential primary would disqualify them from casting ballots in the November general election.

The FCC said that in addition to the civil penalty, Lingo Telecom also agreed to strict rules and requirements for caller ID authentication and that the company would more thoroughly review the accuracy of information provided by its customers and upstream carriers.

“Every one of us deserves to know that the voice on the line is exactly who it says it is,” FCC Chair Jessica Rosenworcel said in a statement. “When AI is used, that must be made clear to every consumer, citizen and voter who comes into contact with it. The FCC will take action when trust in our communications networks is at stake.”

Lingo Telecom did not immediately respond to a request for comment. The company had previously said it strongly disagreed with the FCC’s action, calling it an attempt to impose new rules retroactively.

Kramer, who paid a magician and a self-described “digital nomad” to make the recording, said earlier this year that he was not trying to influence the outcome of the primaries but rather wanted to highlight the potential dangers of AI and push lawmakers to take action.

If found guilty, Kramer could face up to seven years in prison for voter suppression and up to one year in prison for impersonating a candidate.

The nonprofit consumer advocacy group Public Citizen praised the FCC for its action. Robert Weissman, the co-president, said Rosenworcel had “got it right” in saying that consumers have a right to know when they’re getting authentic content and when they’re getting AI-generated deepfakes. Weissman said the case illustrates how such deepfakes “pose an existential threat to our democracy.”

Loyaan Egal, head of the FCC’s enforcement bureau, said the combination of caller ID spoofing and generative AI voice cloning technology posed a significant threat, “both from domestic agents seeking political advantage and from sophisticated foreign adversaries exerting malign influence or engaging in election interference.”