"Dangerous" AI models are coming no matter what
AI models with advanced hacking capabilities will soon be the norm.
The race for advanced AI has hit a regulatory wall as the U.S. government moves to restrict models capable of sophisticated cyberattacks.
The short version
- Anthropic took Claude Fable 5 and Mythos 5 offline following a U.S. export-control directive.
- The government believes guardrails on Fable 5 can be bypassed to access dangerous hacking capabilities.
- Experts argue that restricting a single company is futile as similar capabilities are becoming an industry standard.
What happened
Following a directive from the Trump administration, Anthropic disabled its Mythos 5 and Claude Fable 5 models to prevent use by foreign nationals. While Fable 5 was released to the public with blocks on cybersecurity and biology, the government views it as a national security risk. Mythos 5 is explicitly designed for "dual use," meaning it can help defenders patch software vulnerabilities or allow malicious actors to exploit them.
Why it matters
This clash highlights a broader technological shift. Experts warn that these capabilities are not unique to Anthropic; OpenAI and open-source developers are likely developing similar tools. Because smaller, open-source models can often match this performance through sophisticated prompting, researchers argue that the focus should shift from specific restrictions to transparent, democratic plans for managing inevitable AI advancements in cybersecurity.
Summary by Nerd News Network. Read the full article at Ars Technica — AI via the links above and below.