Generative AI security firm Portal26 recently announced a new solution to address the excessive computation costs associated with automated agent systems being adopted by enterprises. This product is specifically designed to block exorbitant expenses incurred when autonomous AI agents repeat tasks in unintended ways or exceed their intended scope of activity. The core goal of this announcement is to help enterprises operate AI systems stably without suffering unexpected financial burdens, demonstrating a commitment to minimizing the risks of AI adoption within organizations.
Portal26 stated on the 24th (US local time) that it has officially launched a feature to set computation budgets applicable to individual agents, specific workflows, and even the entire organization. The system is designed to automatically slow down processing speeds as users approach their predefined budget limits, and to pause or completely terminate operations if the limits are exceeded. The computation unit used here is the basic unit required for large language models to understand and generate text, and the cost to be paid increases proportionally as this usage grows.
A company representative warned that multi-step connected automated agents can get caught in recursive loop loops or ask too many questions to the system, causing them to deviate from their intended scope of work. In such situations, computation usage can increase exponentially, and companies may receive unexpected high invoices while simultaneously facing instability in their operating environment. Portal26 emphasized that this tool is the first solution dedicated to managing specific risk factors according to enterprise scale, explaining that it goes beyond a simple restriction mechanism to serve as an integrated management platform.
Arty Raman, CEO, pointed out, "Agent-based AI technology is powerful, but if cost management is not handled properly, it can quickly become expensive and chaotic." He added that Uber Technologies serves as an example reflecting the conflict companies face between the speed of AI adoption and cost predictability. He also explained that the new module will provide customers with transparent visibility into detailed usage status, offering decisive help in significantly reducing unplanned invoices.
This update goes beyond simply limiting usage and includes a feature to visualize in real-time how computational resources are used throughout the entire agent system. It is also equipped with adaptive safety mechanisms that intervene automatically when approaching budget limits, showing a vision of making cost predictability the default value of operations rather than relying on post-hoc financial cleanup. This is a strategy tailored to the current period when challenges faced by enterprises expanding generative AI from experimental stages to actual work environments are becoming more prominent. It is essential for controlling the cost incurred with each call in systems where model calls occur in a chain through multiple tasks.
Portal26 has recently introduced an agent management tool focusing on AI security and business value measurement. Paksy Rajan, Chief Product and AI Officer, evaluated this new product as core infrastructure for "Responsible AI Operations," beyond a simple cost control tool. This venture-backed startup has raised a total of $15 million in funding through two rounds, with major investors including Refinery, Shasta Ventures, and Fusion Fund. As competition among enterprises in generative AI shifts from the laboratory stage to the operational stage, 'cost control' has emerged as a key challenge alongside 'performance.' It is expected that the importance of tools for precisely managing computation usage will grow even further as the spread of AI agents accelerates.