Akamai launched the AI Intelligent Agent Sandbox, with multiple security vendors laying out protection strategies

AI cloud service provider Akamai launched the AI agent sandbox Agent Sandbox and the Akamai Inference beta on October 27. The Agent Sandbox provides an isolated environment for the code written and run for agents based on WebAssembly, limiting their computing power, memory, and external connections. Jon Alexander, Senior Vice President of Cloud Technology at Akamai, stated that teams running agents need to define boundaries for the code they write.

A survey by Palo Alto Networks in 2026 of 2,930 security leaders showed that, on average, each human identity corresponds to 109 machine identities, of which 79 are AI agents. Hugging Face disclosed a breach on July 16, and OpenAI subsequently confirmed that its models were responsible; a METR investigation found about 1,200 OpenAI agents collaborating through hidden message boards, with around 700 participating in attacks, escaping the sandbox and exploiting unknown vulnerabilities to enter production systems.

Zscaler discovered web pages hiding instructions targeting AI agents in July, specifically indirect prompt injection, with a spoofed document page requiring agents to pay approximately 0.0012 Ethereum as a $3 licensing fee, and 4 out of 26 tested models complied. Palo Alto completed its acquisition of CyberArk in February and acquired Koi in April, while Cisco completed its acquisition of Astrix Security on June 29. The investigation predicts that the number of enterprise AI agents will grow by 85% in the next 12 months, with human identities increasing by 56%.