- Anthropic launched Mythos, an AI model with powerful, potentially dangerous capabilities
- Mythos preview shared only with select cybersecurity and tech firms for safety testing
- The AI found high-severity vulnerabilities in major operating systems and browsers
Anthropic on Tuesday (Apr 7) launched a new artificial intelligence (AI) general-purpose language model called 'Mythos', which it claims is too powerful to release into the world. Instead, the US-based AI developer rolled out a preview of its new model to a group of handpicked tech and cybersecurity companies that could help implement safeguards to control the model's most dangerous capabilities.
"We do not plan to make Claude Mythos Preview generally available, but our eventual goal is to enable our users to safely deploy Mythos-class models at scale, for cybersecurity purposes, but also for the myriad other benefits that such highly capable models will bring," Anthropic said in a press release.
"To do so, we need to make progress in developing cybersecurity (and other) safeguards that detect and block the model's most dangerous outputs."
Anthropic said AI models have reached a level of coding capability that can surpass most human coders in finding and exploiting software vulnerabilities. The company claimed that Mythos was able to fund high-severity vulnerabilities in "every major operating system and web browser".
"Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely. The fallout, for economies, public safety, and national security, could be severe," the company stated.
Dangerous Capabilities of Mythos
Anthropic cited an example of Mythos discovering a 27-year-old vulnerability in the open-source operating system OpenBSD that may allow hackers to crash any machine running it. Mythos also found a previously unnoticed flaw in video software that had been tested more than five million times by its creators.
The model also found security issues in the Linux kernel, the software that runs most of the world's servers. The vulnerability allows an attacker to take complete control of the machine.
As a precaution, Anthropic has shared a version of Mythos with cybersecurity companies CrowdStrike and Palo Alto Networks, as well as with Amazon, Apple and Microsoft in a project it dubbed "Glasswing."
After Anthropic's release, OpenAI and other AI giants are expected to release models with similar capabilities, with Anthropic CEO Dario Amodei issuing a chilling warning.
"More powerful models are going to come from us and from others, and so we do need a plan to respond to this," Amodei said in a video released alongside the news.
