Amazon's security research led to export controls on Anthropic's AI models. The White House blocked foreign access after CEO Andy Jassy shared findings.
Amazon's security research reportedly influenced the White House's export control directive on Anthropic's AI models. According to the Wall Street Journal, CEO Andy Jassy discussed security concerns with officials shortly before the directive was issued.
Amazon's research showed that specific prompts could make Anthropic's model provide information useful for cyberattacks. The White House then blocked foreign nationals from accessing the model, which also affected Anthropic's own foreign-born researchers.
Anthropic disputes the characterization as a 'jailbreak,' arguing similar vulnerabilities exist in other public models like GPT-5.5. This highlights tensions between AI safety and export controls, potentially shaping future AI regulation debates.