Akamai: AI Security Moving to the Edge
The future of AI in security isn't just in the cloud, it's at the edge. An Akamai director argues that moving AI inference closer to devices provides critical speed for threat detection and better data privacy for compliance. This hybrid approach also ensures that basic security functions continue even if the cloud connection is lost.
The global push for edge computing is massive, with Gartner projecting 75% of enterprise-generated data will be processed at the edge by 2025, a huge jump from 10% in 2018. Investment is following, with the market expected to nearly double to almost $400 billion by 2028. This architectural shift enhances resilience; edge AI can enable critical infrastructure to operate and defend itself locally even when central cloud connections are severed by a cyberattack. By processing data at the source, AI-powered thermal cameras, for example, can perform complete threat analysis in real-time without cloud dependency. However, this decentralization also creates a broader attack surface, introducing new vulnerabilities across countless devices outside of a central data center. Adversaries are also leveraging AI to scale and enhance their own attack methods, creating a new front in the cybersecurity landscape. In response, Akamai has developed specific tools like its Firewall for AI, which provides multilayered protection at the edge against threats like prompt injection, unauthorized queries, and data scraping. This solution is designed to secure both the inbound queries to AI models and the outbound responses they generate. To build out this edge-native security, Akamai has formed strategic partnerships, including a collaboration with Aqua Security to embed runtime protection directly within AI containers. It has also worked with NVIDIA to enable agentless microsegmentation that can enforce Zero Trust policies directly at the edge without disrupting operations. For defenders, the primary advantage of edge AI is the ability to move from reactive to proactive security. Machine learning algorithms can analyze network traffic and system behavior on-site to flag anomalies, identify potential zero-day exploits, and autonomously isolate compromised devices to mitigate threats in real time. This evolving landscape makes a Zero Trust security model essential. Instead of trusting devices automatically, this approach requires continuous verification for every user, device, and application. For penetration testers, understanding how to test and bypass these distributed, AI-hardened environments is becoming a critical skill.
