Anthropic collaborates with tech giants such as Microsoft Corporation (MSFT.US) to test new AI models to address cybersecurity risks.

Anthropic announced on Tuesday that it will launch a new initiative in collaboration with several tech giants to proactively address potential cybersecurity threats posed by more powerful AI models. According to reports, Anthropic is launching a collaborative project called "Project Glasswing" and will make an unreleased AI model called "Mythos" available to multiple tech companies including Amazon.com, Inc., Apple Inc., Microsoft Corporation, and Cisco Systems, Inc. These companies will utilize the model to identify security vulnerabilities in their own products and share research findings with the industry. Anthropic stated that there are currently no plans to make Mythos available to the public, but in the future, they will enhance the model's security "fence" based on feedback from the Glasswing project. This move reflects the industry's growing concerns about advanced AI models being abused by hackers or state-sponsored attackers. It is widely believed in the industry that as AI capabilities increase, its ability to analyze code and discover vulnerabilities also strengthens, potentially enabling it to bypass existing network defense systems. Anthropic's competitor, OpenAI, has previously emphasized similar risks and launched pilot projects to prioritize providing AI tools to defense users. Newton Cheng, head of Anthropic's cybersecurity team, stated that this is not just a problem for one company but a challenge that the entire industry and government need to address together. "We hope that through Glasswing, defense entities can gain an advantage before the technology spreads." The company also revealed that they have been in communication with relevant U.S. government agencies regarding Mythos' security capabilities and have partnered with the Cybersecurity and Infrastructure Security Agency (CISA) and the National Institute of Standards and Technology (NIST). Although Mythos was not specifically developed for cybersecurity, its performance has garnered attention. Anthropic stated that the model has discovered several critical vulnerabilities, including a key internet software flaw that has existed for 27 years and a gaming code defect that went undetected for 16 years, despite being scanned over 5 million times by automated testing tools. To prevent technological abuse, Anthropic stated that they have implemented strict access controls for participants in the Glasswing project but did not disclose specific details.