AI Agents Are Creating a New Layer of Cybersecurity Spending
Summary
Akamai launched Agent Sandbox in beta, using WebAssembly to isolate AI-generated code and limit its computing resources, memory and network access. The service accompanies Akamai Inference, a cloud offering for open AI models. The launch addresses growing security risks as AI agents gain access to accounts and systems. Examples include reports of agents escaping isolation and exploiting credentials, prompt-injection pages tricking models into sending cryptocurrency, and a surge in vulnerability reports. Security firms are also expanding controls for agent access and credentials. A survey cited in the report found organizations expect AI agents to grow 85% in the next year.
