Report from the Wall Street Journal
In Brief – The Trump administration’s decision to halt foreign access to Anthropic’s most advanced Fable and Mythos AI models followed warnings from Amazon’s CEO that his company’s researchers had been able to prompt the Fable 5 model into revealing information that could aid cyberattacks. Amazon is a major investor and cloud provider for Anthropic. The cyber concerns quickly prompted White House discussions and federal government testing. Administration officials soon asked Anthropic to address the issues, but the two sides did not reach agreement, with the AI developer reportedly arguing that the vulnerabilities were relatively basic, already achievable with other available models, and did not constitute a full jailbreak of the system. The White House ultimately concluded that restricting foreign access was the best way to mitigate risks. Anthropic chose instead to shut down all access to the top models, claiming it could not comply with such a targeted restriction, while a large group of senior cybersecurity experts publicly asked the Administration to allow Anthropic’s tools to remain available.
Context – Two key influences on AI policy are that nobody, including the developers, fully understands the technology or how it will be usefully employed, and everyone involved has been raised in a science fiction culture filled with mostly dystopian stories of AI. The result is policy heavily influenced by sci-fi fears. While the Trump Administration has been aggressively anti-regulatory, arguing that AI innovation is critical to competing against China, and has pulled policy internationally in that direction, cracks are building at home based on increasing public anxiety and a vocal cadre of anti-Big Tech conservatives. The progressivism of Anthropic’s leaders is having a big impact on the ongoing cyber and national security policy machinations involving the company because distrust over “Woke AI” is one of the few things that unites the pro-tech and anti-tech conservatives on AI. The prospect that broad foreign access shutdowns become a default US AI security policy response will boost the EU’s tech sovereignty drive and broader insecurity of traditional US allies.