Home / Technology / US AI Export Ban: A Global AI Rulebook Test
US AI Export Ban: A Global AI Rulebook Test
20 Jun
Summary
- US orders Anthropic to restrict powerful AI models export.
- AI export controls face first real test amidst national security concerns.
- Past US export control attempts on encryption and spyware had mixed success.
Last Friday, the White House directed Anthropic to halt the export of its advanced AI models, Fable and Mythos, due to unspecified national security concerns. This directive also restricts access for foreign nationals within the United States, leading to the immediate unavailability of both models for over a week.
This situation represents a critical trial for U.S. export controls aimed at managing frontier AI technology, mirroring past efforts with encryption and spyware that saw limited success. The resolution of this standoff could significantly influence Anthropic's international market access and set precedents for other AI laboratories.
Anthropic had previously marketed Mythos as a potentially destructive cyber tool, limiting its access to about 150 vetted entities. However, recent events, including suspected ties to China via a South Korean telecom partner and alleged circumvention of Fable 5's safeguards by Amazon researchers, triggered the U.S. government's action.
The Commerce Department issued an export control directive, compelling Anthropic to limit access to its products within approximately 90 minutes. This mirrors historical government attempts to control dangerous cyber technology, such as the "Crypto Wars" in the 1990s surrounding encryption software PGP.
Decades ago, similar U.S. efforts to control encryption and later, spyware through agreements like the Wassenaar Arrangement, faced significant challenges. Weaknesses included non-adherence by certain countries and discretionary application by others, allowing companies like Hacking Team to export surveillance tools widely.
Despite some successes, like the shutdown of FinFisher, many spyware makers have relocated to countries with laxer controls or operate within regions that have been slow to curb exports to authoritarian regimes. The current impasse between Anthropic and the U.S. administration may result in either a push for global AI competitiveness or a significant compliance burden for U.S. AI firms serving foreign clients.
Given historical precedents, government-mandated export controls may prove insufficient to prevent the misuse of powerful dual-use cyber technologies by malicious actors.
