American artificial intelligence company Anthropic is dialing back on the full release of its new AI model, Claude Mythos Preview, after revealing it is too dangerous for public usage, with U.S. Treasury Secretary Scott Bessent issuing warnings, The Hill reports.
As part of Anthropic’s latest initiative, Project Glasswing, Claude Mythos Preview focuses on identifying and patching security vulnerabilities in critical software. While available only to a select group of technology firms, including Microsoft, Apple, CrowdStrike, and Amazon Web Services, as well as 40 other organizations, Anthropic said the initiative was formed after discovering the capabilities of Mythos Preview, stating that the model “could reshape cybersecurity.”
Mythos Preview has already found several high-security vulnerabilities, including some in each major operating system and web browser, that were unknown to the software’s developers prior to this, with some dating back over two decades. Before models like Mythos, these vulnerabilities would have gone undetected for years.
Currently, the technology is enabling hackers and foreign adversaries to detect these vulnerabilities with ease. “Given the rate of AI progress, it will not be longer before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely,” Anthropic wrote in its announcement.
“The fallout — for economics, public safety, and national security — could be severe.”
Forbes outlined some findings, such as a vulnerability in OpenBSD, an operating system known as one of the most secure in the world, used to run firewalls and substantial infrastructure. It found a bug permitting anyone to crash a machine remotely just by connecting to it.
After 27 years, human review missed it.
It also identified a 17-year-old remote code execution vulnerability in FreeBSD, giving anyone access to a machine running from anywhere on the internet.
While highlighting the dangers of these capabilities, the company also emphasizes the opportunity to find and fix software flaws and to build new software more easily with fewer security bugs.
The risk spooked Bessent and Federal Reserve Chair Jerome Powell enough to call an urgent meeting with the CEOs of leading financial institutions to warn of cyber risks posed by Anthropic’s latest AI model, according to Reuters
.The April 7 meeting in Washington, D.C., sought to ensure that banks are aware of the risks posed by Mythos and similar models and are taking appropriate steps to defend their systems.
RELATED CONTENT: Artificial Intelligence Is Changing The Way Students Pick College Majors