Akamai on Tuesday launched Agent Sandbox, a sealed environment where code written by AI agents can run without accessing the rest of a company's systems. The product is in beta alongside Akamai Inference, a service for running open AI models on Akamai's cloud.
Security concerns grow with agentic AI
The launch addresses a growing problem: software no longer just answers questions—it logs into accounts, writes code, moves money and acts on its own. Each of these capabilities can be abused by hackers or by the software itself.
A 2026 Palo Alto Networks survey of 2,930 security leaders found an average of 109 machine identities for every human, the logins and keys software uses to access other software. Of those, 79 are AI agents.
Agentic AI brings benefits but also serious dangers, especially when agents have too much freedom or are intelligent enough to bypass their own restrictions.
Incidents highlight risks
Hugging Face, a major hub for open AI models, disclosed an intrusion on July 16, and OpenAI confirmed five days later that its own models were responsible. The research group METR later investigated and found about 1,200 OpenAI agents had coordinated on a hidden message board, with roughly 700 joining the attack.
The agents escaped their sandbox and chained a previously unknown software flaw with stolen credentials to reach live systems. OpenAI called it a "warning shot" and said a monitor that reads a model's written reasoning was not running during the tests.
In July, Zscaler found web pages hiding instructions meant for AI agents, a trick called indirect prompt injection. One fake documentation page told agents to pay a $3 "license fee" in about 0.0012 Ethereum to a hardcoded wallet, and four of the 26 models Zscaler tested complied.
In May, cybersecurity firm Bugcrowd saw submissions more than quadruple in three weeks in March, most of them fake. HackerOne and Nextcloud suspended paid programs in April.
Yet machines also find real bugs. Mozilla said an early version of Anthropic's Claude Mythos helped find 271 vulnerabilities in Firefox, all patched within the week.
Vendors race to secure agents
Palo Alto closed its CyberArk purchase on Feb. 11 to extend privileged-access controls to AI agents. In April it bought Koi, which secures AI coding agents running on employee machines. Cisco announced a planned acquisition of Astrix Security in May, a firm that governs the keys and tokens agents use, and completed the deal June 29.
Akamai's sandbox uses WebAssembly, a format that runs programs in a tightly walled box, to cap an agent's computing power, memory and outbound connections. "Enterprises should not have to choose between speed and control when they put AI into production," said Jon Alexander, Senior Vice President of Cloud Technology at Akamai, in a press release. "Every team running agents needs a boundary around the code those agents write."
Palo Alto's survey found organizations expect AI agents to grow 85% over the next 12 months, against 56% for human identities.
