Google Security Leader Shares Advice for Securing AI Systems
Summary
Google’s James Brodsky recommends securing AI end to end—from infrastructure and data pipelines to model behavior—and highlights access controls, DLP, monitoring, threat-intelligence partnerships and ongoing staff learning.
Key points
- Brodsky says AI security should cover underlying infrastructure and services, data sources and pipelines, and model inputs and outputs.
- He highlights monitoring for model poisoning and tampering, with visibility into technical infrastructure, developer behavior and access controls.
- The article recommends starting with data-location awareness, access controls and data loss prevention before adopting more complex measures.
- It advises coordinating with HR and considering legal and regulatory frameworks when monitoring employee behavior.
- The article recommends combining security platforms with threat intelligence and investing in continuous team learning, including through Google AI Essentials courses and threat-intelligence CTFs.
Article Details
- Topic
- Framework and practical guidance for securing AI systems, including layered protection, monitoring for model poisoning and prompt injection, strategic partnerships, and security-team development.
People
Vendors
GoogleI recently had a conversation with James Brodsky, Head of Global Security Architects at Google, on The Future of ThreatIntelligence Podcast, where we tackled a challenge that's keeping security leaders up at night: howTeam Cymrureflects a broader industry trend we're seeing, where combining specialized expertise - like Team Cymru's external threat intelligence capabilities fused with Google's SecOps platform - this creates amore