Google has unveiled its new AI Cyber Defense Initiative, aimed at enhancing digital security using artificial intelligence (AI). The initiative includes the open-sourcing of Magika, an AI-powered tool that is already being used to protect Gmail users from potentially harmful content.
The announcement was made ahead of the Munich Security Conference, where Google will launch the AI Cyber Defense Initiative on February 16. The company believes that its expertise in deploying AI can help address the defender's dilemma, which refers to the challenge faced by defenders who must constantly defend against new threats while attackers only need to exploit a single vulnerability.
Google will release a report titled 'Secure, Empower, Advance: How AI Can Reverse the Defender's Dilemma' during the conference. The report will outline the proposed policy and technology agenda, emphasizing how AI can enable security professionals to scale their work in threat detection, malware analysis, vulnerability detection, vulnerability fixing, and incident response.
One of the tools already utilizing AI is RETVec, a multilingual neuro-based text processing model used by Gmail to improve spam detection rates by 40% and reduce false positives by 19%. As spam often carries malicious payloads, this improvement is significant.
Another AI-powered tool called Magika is used by Gmail to identify file types and help defenders detect malware. Google plans to open-source Magika, which is also employed to protect Google Drive and the Safe Browsing functionality. Magika has demonstrated superior performance compared to conventional file identification methods, achieving up to 30% higher accuracy. For content that is particularly challenging to identify, such as JavaScript, Powershell, or VBA, the precision rate reaches an impressive 95%.
As part of the AI Cyber Defense Initiative, Google will invest in AI-ready infrastructure and release new tools, research, and security training for defenders. The company will also launch a 'Secure AI Framework' to facilitate collaboration on AI system security best practices. Additionally, Google for Startups Growth Academy's AI for Cybersecurity Program will support 17 startups across the U.S., U.K., and Europe, fostering a transatlantic ecosystem for cybersecurity that incorporates AI tools and the necessary skills.
To further advance AI-powered security, Google is providing $2 million in research grants. The funding will support projects focused on code verification enhancement and the development of large language models that are more resilient to threats. Researchers at the University of Chicago, Carnegie Mellon, and Stanford will be the initial beneficiaries of these grants.
