Acalvio Technologies has unveiled Deception Guardrails, a new security capability designed to protect agentic AI systems by detecting and disrupting compromised AI agents before they can access enterprise assets. The company said the launch addresses a gap left by traditional AI guardrails, which primarily monitor inputs and outputs but offer limited visibility into agent behavior after a compromise.

As organizations increasingly deploy AI agents capable of autonomous reasoning and task execution through tools and APIs, Acalvio said the enterprise attack surface has expanded. The company noted that compromised AI agents can exploit credentials, tools and external systems in ways that bypass conventional guardrails.

The company said its patent-pending Deception Guardrails shifts AI security from reactive filtering to preemptive defense by embedding deceptive assets that detect malicious activity, feed attackers fabricated data and alert security operations teams before production systems are affected.

"Reactive guardrails are designed to keep well-behaved AI systems on the road, but they do nothing to stop a hijacked agent driven by a malicious actor," said Ram Varadarajan, CEO at Acalvio. "With our patent-pending Deception Guardrails, we are moving the industry from reactive filtering to preemptive defense. If an AI agent goes rogue or its infrastructure is manipulated, our deceptive assets rapidly detect the misalignment, feed the attacker fabricated data, and alert the SOC before real enterprise assets are compromised."

Acalvio also pointed to a recent Hugging Face incident, which prompted a briefing from cyber authorities representing the Cloud Security Alliance, SANS and RSAC. According to the company, the briefing recommended using deception technologies because AI agents cannot easily distinguish legitimate credentials and systems from deceptive ones, encouraging organizations to deploy fake identities, credentials, package registries, datasets, APIs and clusters to slow attackers and generate high-confidence indicators.

Lawrence Pingree, Head of Research at Software Analyst Cyber Research, said the approach builds on deception technologies already used to detect sophisticated attackers in enterprise environments while extending those capabilities to AI systems.

"Acalvio has spent years perfecting deception technologies to detect sophisticated attackers inside enterprise environments. Extending those same principles to AI agents is a natural and compelling evolution," Pingree said. "By embedding deceptive assets directly into agent workflows and surrounding AI infrastructure with decoys, organizations gain a powerful new layer of detection that complements existing AI safety and governance controls."

Acalvio said Deception Guardrails extends its ShadowPlex platform with capabilities including enterprise-wide honeytokens and decoys across cloud and on-premises environments, deceptive assets embedded into AI agent workflows, decoy AI infrastructure including Model Context Protocol servers, Retrieval-Augmented Generation systems and AI agents, as well as real-time detection of malicious manipulation, jailbreak behavior and prompt injection attempts.

The company said the technology is intended to provide early warnings that help security teams identify and respond to compromised AI agents before business-critical systems are affected while supporting enterprise-scale AI deployments.

Acalvio will demonstrate Deception Guardrails during Black Hat USA at Booth #8606 in the AI Zone.