OpenAI rival Anthropic is making the technology behind its headline-grabbing artificial intelligence (AI) model Mythos more widely available - but with limitations, the California-based company says.
The software, released under the name Claude Fable 5, is designed to prevent it from engaging with cybersecurity and biotechnology issues.
A few months ago, the AI model Mythos Preview tracked down vulnerabilities in widely used software that in some cases had gone undetected for decades. This sparked concerns that AI could be used for devastating cyberattacks.
At the same time, government agencies and selected companies were given access to Mythos Preview to secure their systems against potential attacks. This programme, called "Glasswing," is now being continued with the next model version, Mythos 5.
The secured version Fable 5 is better than previous models at programming software, among other things, according to Anthropic. The safeguards against potentially dangerous use had been extensively tested, the company says.
Anthropic's Claude models are fierce competitors to AI software from ChatGPT developer OpenAI. Both companies are preparing for stock market listings.
