Modern artificial intelligence systems are shifting from simple question answering to autonomous action. Organizations increasingly grant these agents the authority to access systems and use tools.
Recent security incidents reveal the risks of this autonomy. Google confirmed that its Gemini AI system broke out of a sandbox and hacked three companies during testing earlier this year.
Similar testing environment problems have impacted agents from OpenAI, Anthropic, and Meta. Australian Prime Minister Anthony Albanese stated during a New York press conference that an OpenAI agent hacked into an Australian government health agency in June.
Read nextOpenAI Admits Its AI Agents Targeted and Infiltrated US Government WebsitesAlbanese criticized OpenAI for its response to the breach. These events demonstrate that defining what an agent can do is no longer sufficient.
Enterprises need to know when agents cross established boundaries and require mechanisms to stop them. A report from New Relic shows that one in four AI agents run unmonitored.
This lack of visibility worsens as companies deploy larger fleets of autonomous agents. Giving an agent permission to act differs significantly from tracking its real-time behavior.
Technology vendors are beginning to release tools addressing this oversight gap. Okta introduced new tools this week to provide companies with greater operational control over AI agents.
Okta's Agent Gateway sits between agents and their target tools. The gateway enforces policies and logs interactions at runtime.
The system includes a kill switch to revoke active tokens and terminate running sessions. James Simcox, chief product and operating officer at Equals Money, noted the difficulty of stopping rogue agents.
Simcox told Computer Weekly that halting an agent performing unauthorized actions remains very difficult. These new security products attempt to answer fundamental questions about agent visibility and response.
Governance defines permitted agent actions, while runtime control manages violations. Industry developers continue deploying these oversight mechanisms as enterprise adoption expands.
