Per Silicon Angle, Pretxt AI has launched with $5 million in seed funding to address the risks associated with deploying autonomous agents in enterprise production environments.
The startup introduces a runtime supervision layer designed to monitor AI agents as they execute tasks. By comparing agent behavior against corporate policies in real time, the platform intervenes if an agent drifts off-course, preventing potential business errors. Unlike traditional security tools that rely on secondary large language models—which can introduce latency and costs—Pretxt utilizes a near-deterministic engine to evaluate actions in milliseconds. This approach is reportedly 60 times faster than competing methods. Co-founders Brian Soby and Aaron Costello, who previously worked together at AppOmni Inc., developed the tool to help organizations safely scale AI projects in regulated sectors like finance, insurance and healthcare. The platform allows teams to define guardrails in plain language, ensuring agents follow specific workflows without requiring deep technical expertise. By halting agents during unforeseen scenarios and escalating them to human decision-makers, Pretxt aims to bridge critical gaps in risk management. The tool is currently available in early access, offering a proactive alternative to reactive security measures that typically flag errors only after damage has occurred.
Source: Silicon Angle
