tech
Amazon researchers used Anthropic AI to find cybersecurity weaknesses
Researchers at Amazon were reportedly behind the jailbreak that prompted the Trump administration to impose export restrictions on Anthropic’s most advanced artificial intelligence models, according to the Wall Street Journal.
TL;DR
- Amazon researchers are reportedly behind a jailbreak of Anthropic's AI models.
- The jailbreak prompted the Trump administration to impose export restrictions on Anthropic's Fable 5 and Mythos 5 models.
- Anthropic has blocked public access to these models in compliance with the Commerce Department's directive.
- Company executives and government officials reportedly met shortly after the jailbreak.
- Experts argue that using AI to bypass guardrails is essential for cybersecurity defense, comparing it to offense.
- One expert described the administration's response as a "regulatory overreaction" and a "misread" of the research.
- Amazon acknowledged consulting with the government on cybersecurity risks but did not confirm performing the specific jailbreak testing.
- Anthropic believes the government's decision was tied to the jailbreak but contends the vulnerabilities found were minor and not new.