…Concerns over cyberattack vulnerabilities prompt government restrictions on foreign access
The United States government’s decision to restrict foreign access to some of Anthropic’s most advanced artificial intelligence models was reportedly triggered by security concerns raised by Andy Jassy, Amazon’s Chief Executive Officer.
According to a report by The Wall Street Journal, Jassy alerted senior US officials, including Treasury Secretary Scott Bessent, after Amazon researchers discovered that Anthropic’s Fable 5 model could be manipulated into revealing information that might facilitate cyberattacks.
The concerns prompted a review by US authorities and ultimately led to restrictions preventing foreign governments, businesses and individuals from accessing the advanced AI models.
Sources familiar with the matter told the Wall Street Journal that Amazon researchers conducted tests on the Fable 5 model using specially crafted prompts and found ways to bypass certain safeguards designed to prevent the disclosure of sensitive information.
The findings reportedly prompted White House officials to convene discussions on potential security risks associated with the model.
Independent security researchers were subsequently asked to verify Amazon’s claims, while US authorities engaged Anthropic to address the identified vulnerabilities.
According to the report, officials requested that Anthropic either resolve the security issues or temporarily withdraw the model from public access.
After reviewing the situation, the administration concluded that limiting foreign access to the technology represented the most effective way to mitigate potential risks.
The report stated that President Donald Trump approved the restrictions despite concerns within parts of the administration that such measures could slow innovation in the rapidly evolving AI sector.
The decision was reportedly reinforced following discussions between government officials and Anthropic CEO Dario Amodei.
Some administration officials claimed the company appeared reluctant to collaborate with government security experts in addressing the vulnerabilities, further deepening concerns about its ability to manage the risks associated with increasingly powerful AI systems.
Sources cited by the Wall Street Journal said scepticism about Anthropic’s security oversight had existed within parts of the administration even before the latest controversy.
In response to the directive, Anthropic announced that it would disable access to both its Mythos and Fable models globally to ensure compliance with government requirements.
The company warned that the move could affect organisations that use the models to identify software vulnerabilities and strengthen cybersecurity systems.
Anthropic also noted that many of its researchers are foreign-born, meaning the restrictions could limit their ability to work on the company’s most advanced AI models.
Anthropic has challenged the technical justification for the government’s action.
The company argued that the vulnerability identified by Amazon represented a limited and highly specific “jailbreak” technique that exposed only minor issues already known within the AI research community.
According to Anthropic, similar capabilities are available in other publicly accessible AI systems, including those developed by OpenAI.
The company maintains that the issue does not justify broad restrictions on access to its technology.
The development highlights growing tensions between rapid AI innovation and national security concerns.
Anthropic has previously advocated stronger safeguards for advanced AI development, calling for industry-wide mechanisms that would allow developers to slow or pause the deployment of increasingly powerful systems if they begin evolving beyond safe levels of human oversight.
The latest dispute underscores the challenges governments and technology companies face as artificial intelligence becomes more capable, more influential, and potentially more difficult to regulate.
