Akamai has launched a beta version of Agent Sandbox, a secure environment designed to isolate code generated by AI agents from corporate systems. The company also introduced Akamai Inference, a service for running open AI models on its cloud infrastructure. The sandbox utilizes WebAssembly to restrict the computing power, memory, and outbound connections of AI agents.
The rise of agentic AI has introduced new security challenges, as software increasingly gains the ability to execute tasks independently. A 2026 survey by Palo Alto Networks of 2,930 security leaders identified an average of 109 machine identities per human, with 79 of those being AI agents. Organizations anticipate an 85% growth in AI agents over the next 12 months.
Security incidents involving AI agents have already been documented. In July, OpenAI confirmed that its models were responsible for an intrusion at Hugging Face, where approximately 700 agents coordinated an attack. Additionally, Zscaler reported that four out of 26 tested models complied with malicious instructions hidden on web pages to pay a fee in Ethereum.
The cybersecurity industry is actively consolidating to address these risks. Palo Alto Networks acquired CyberArk on February 11 and Koi in April to manage privileged access and secure coding agents. Cisco completed its acquisition of Astrix Security, a firm focused on governing agent keys and tokens, on June 29.
