Artificial Intelligence & Machine Learning , Next-Generation Technologies & Secure Development
Google Calls for Framework Giving AI Access to Incident Data
Computing Giant Launches 'AI Cyber Defense Initiative'Computing giant Google called on governments across the globe to create a cross-border framework to ensure that artificial intelligence can effectively fight cyberthreats. The company said the technology could offset the inherent advantages attackers have had in cyberspace since almost the start of the internet.
See Also: Accelerate innovation with Copilot+PCs- fastest, most intelligent Windows PC
In a paper presented in the lead-up to the annual Munich Security Conference, Google said "AI affords the best opportunity to upend the Defender's Dilemma," a reference to this aphorism: "Defenders have to be right every time. Attackers only need to be right once."
AI can be used in "two fundamental paradigm shifts" that address the root causes of cybersecurity, the company said in the launch of what it calls its AI Cyber Defense Initiative. AI can analyze diverse datasets far larger than any human can handle, and it can enhance defense, eventually moving from being an assistive technology to an autonomous one, Google said.
Getting there will require an "international framework that preserves the ability of AI systems to learn from global incident data and operate across borders," the company said, as part of a "road map for reversing the defender's dilemma."
Cyber defenders for years have pined for a comprehensive repository of incidents, but that goal has been elusive amid concerns about corporate liability, reputation and the security vulnerability potentially created by having such a repository. The U.S. federal government is developing a rule requiring some operators of critical infrastructure to report substantial cyber incidents within 72 hours, but the reports will likely fall short of Google's vision of having access to "global incident data."
Enabling AI to learn from incident data would require controlling access to incident data so that only defenders have access to specialized training data sets, the company said. Still, "this is our once-in-a-generation moment to change the dynamics of cyberspace for the better," it said.
Google called for ensuring that AI systems follow "secure by design" principles. It said that paying attention to AI model risks such as data poisoning overlooks the more straightforward way to attack models: through vulnerabilities in the hardware and software that underpin the technology.
The Mountain View giant also endorsed more scientific research, such as new AI-driven techniques for vulnerability discovery. In the future, scientific breakthroughs could create AI agents for security - autonomous entities that could manage cybersecurity issues. "How to build these agents, measure and monitor their performance and accuracy, and explain their actions is an active research area," Google said.
As part of its AI Cyber Defense Initiative campaign, Google said it will open-source its AI-powered file type identification tool Magika to help defenders detect malware. It also announced a new class of 17 companies in the United States, the United Kingdom and Europe that will join its Google for Startups Growth Academy's AI for Cybersecurity Program, a three-month program for AI startups that aims to teach them to "grow and innovate responsibly."