By Bo Erickson and Jessica Koscielniak
WASHINGTON, April 17 (Reuters) – The Trump administration and Anthropic’s CEO on Friday discussed working together for the first time since a dispute earlier this year between the Pentagon and the AI firm over how that company’s models should be used.
The meeting, suggesting the two sides might be on a path to rebuilding trust, took place amid growing fears the AI startup’s latest model will supercharge cyberattacks.
The Trump administration, central bankers across the globe and industries are racing to get up to speed on Anthropic’s model Mythos and its ability to make complex cyberattacks both easier and quicker to execute.
The banking industry, with its legacy technology systems, is particularly vulnerable. Government officials in at least three countries – the U.S., Canada and Britain – have met with top banking officials to discuss the threats posed by Mythos.
“We discussed opportunities for collaboration, as well as shared approaches and protocols to address the challenges associated with scaling this technology,” the White House said in a statement that described the meeting with Anthropic as “productive and constructive.”
The U.S. government and the Silicon Valley firm have been working through a dispute over how Anthropic’s AI should be used. The disagreement came to a head in late February when U.S. President Donald Trump directed the government to stop using Anthropic, and the Pentagon said it would declare the startup a supply-chain risk.
“The conversation also explored the balance between advancing innovation and ensuring safety. We look forward to continuing this dialogue and will host similar discussions with other leading AI companies,” the White House statement said.
Anthropic did not immediately respond to a request for comment.
Announced on April 7, Mythos is first being deployed to a select group of companies as part of Anthropic’s “Project Glasswing,” a controlled initiative under which the organizations are permitted to use the unreleased Claude Mythos Preview model to search for cybersecurity vulnerabilities.
(Reporting by Bo Erickson and Jessica Koscielniak; Writing by Doina Chiacu; Editing by Caitlin Webber and Rosalba O’Brien)
