Anthropic is clamping down on loopholes that have allowed Chinese firms, such as Ant Group, to access its artificial intelligence models despite existing restrictions. According to sources cited by the Financial Times, engineers have been bypassing controls through cloud providers and overseas subsidiaries.
The move highlights the growing challenge of enforcing AI technology export controls. Even as companies like Anthropic implement stringent policies, users have found creative workarounds to continue accessing powerful models from restricted regions.
The Financial Times reports that Anthropic has identified specific patterns of unauthorized access and is now closing those gaps. The company has not disclosed which specific cloud providers or subsidiaries were involved in the workarounds.
This tightening comes amid broader US government efforts to restrict Chinese access to advanced AI technologies. Anthropic's actions could set a precedent for how other AI developers handle geographic access restrictions.
Some experts argue that completely preventing access may be unrealistic, as motivated engineers will continue finding new methods to circumvent controls. The cat-and-mouse dynamic between access restrictions and workarounds is likely to persist.