▲ Anthropic
U.S. artificial intelligence developer Anthropic dispatched a delegation of its top technical experts to Washington over the weekend to negotiate AI export controls with the Donald Trump administration, The Wall Street Journal (WSJ) reported on June 14, citing internal sources.
According to the WSJ, Anthropic is reportedly seeking to coordinate a compromise regarding the Trump administration's total ban on the use of its latest AI models, "Fable 5" and "Mythos 5," by foreign governments, organizations, and individuals.
Sources stated that both Anthropic and the White House are hoping to find a reasonable middle ground.
The WSJ noted that some government officials view the face-to-face coordination between the Anthropic delegation and the U.S. government, conducted in the form of a joint working group, as a significant step toward a compromise.
On Sunday, June 14, Anthropic executives held frequent emergency conference calls with senior administration officials, including Secretary of Commerce Howard Lutnick and National Cyber Director Sean Cairncross, to discuss the export controls.
It is reported that Tom Brown, co-founder and Chief Computational Officer (CCO) of Anthropic, and Sarah Heck, Head of Public Policy, personally participated in these meetings to explain the company's technical safety measures.
Anthropic previously stated that it has faced various emergency situations due to the government's decision, including the temporary suspension of global services for Fable 5 and Mythos 5, as well as restricting access to these models even for its own foreign national employees.
The two AI models subject to the export controls are products derived from "Mythos," a high-performance AI recently developed by Anthropic.
Mythos caused a stir when it was revealed that the AI could detect zero-day vulnerabilities (undisclosed security flaws) and generate attack tools at a speed surpassing that of human hackers, raising concerns that it could be misused for large-scale automated cyberattacks.
Mythos 5 is a product permitted only for a limited number of institutional clients, such as computing infrastructure firms, while Fable 5 is a version with sophisticated performance restrictions (guardrails) applied so that the general public can use it without the risk of it being misused as a hacking tool.
The Trump administration reportedly decided on the export controls based on the view that there is still a risk that specific groups or individuals could disable Fable 5's guardrails and use it as a weapon for hacking or terrorism.
Anthropic maintains the position that these concerns stem from technical misunderstandings and that its guardrails are practically sufficient.
Earlier this year, Anthropic experienced an unprecedented conflict when it was designated a "supply chain risk company" by the U.S. Department of Defense after urging that its AI used by the U.S. military not be utilized as autonomous lethal weapons.
However, after the overwhelming performance of Mythos caused a ripple effect, the conflict between Anthropic and the Trump administration entered a lull, and the company has been in close discussions with the White House regarding pending issues such as AI cybersecurity.
(Photo: Yonhap News)
※ Please note: This article was translated by AI and may contain errors.