GET DAILY REPORT FREE
U.S. Commerce officials have begun applying export-control-like restrictions to the most advanced AI models, treating them in a similar way to sensitive dual-use technologies, such as high-end chips or cryptographic systems, Bloomberg reports.
Under this approach, companies developing frontier models may be required to seek government approval before allowing certain users, especially foreign nationals, to access or use those systems.
Anthropic’s most capable models became subject to these restrictions after national security concerns were raised.
The government’s concern is that highly advanced models could potentially be misused if their safety protections are bypassed, including through so-called “jailbreak” techniques that trick models into ignoring safeguards. Officials view this as a risk for cyber operations, automated exploitation or other high-impact misuse.
A key practical issue is enforcement. Anthropic reportedly does not have a clean technical way to reliably separate or restrict usage based on nationality in real time at the model level.
Because of this limitation, complying with the requirement to control foreign access effectively forced the company to restrict or disable access more broadly, impacting general availability of its top-tier systems rather than just narrowly limiting certain users.
Anthropic has pushed back on the severity and framing of the security concerns, arguing that the cited vulnerabilities are not unique to its systems and that similar jailbreak-related risks exist across all leading frontier models. The company’s position is that overly strict controls could end up constraining legitimate research and commercial use of advanced AI without clearly improving safety.
