AI/ML , Governance, Risk and Compliance , Application security AI plans are being approved. Recovery plans are not July 27, 2026 Share By Paul Wagenseil Credit: Adobe Stock Images AI agents promise to automate routine work, speed up business processes and improve operational efficiency. Yet many organizations find that managing those agents requires more effort than expected. According to Rubrik Zero Labs , 81% of IT and security leaders say that AI agents result in a net loss of efficiency because the time spent auditing and monitoring the agents exceeds the amount of time they were intended to save. It's especially burdensome for operations, security and governance teams who make sure that autonomous systems behave safely. "The efficiency case is not playing out as expected ," observes Rubrik Chief Transformation Officer Kavitha Mariappan. This doesn't mean organizations should abandon AI agents . But it does mean they need to rethink how those agents are deployed, and how those agents and their work are recovered when things go wrong. Why organizations aren't gaining efficiency from AI agents Many organizations deploy AI agents because competitive pressure rewards speed of innovation. Unfortunately, the governance infrastructure around AI agents has not kept pace. As production ramps up, security and operations teams absorb the unexpected costs of continuously auditing agent behavior, validating actions and managing risk. "The costs are being shifted to teams responsible for maintaining operations," Mariappan says. "Because organizations are deploying agents faster than they can build the necessary governance infrastructure, the operational burden is high, often undermining the expected ROI of the efficiency gains." Gartner predicts that 40% of existing AI agents will be decommissioned by 2027, chiefly because organizations will have deployed them faster than they could govern them. " It's unfortunate," observes Mariappan. "They are effectively racing to innovate without first establishing the necessary governance to handle the agent-induced operational strain." Why the difficulty of rolling back AI-agent actions slows down organizations The governance challenge becomes even more apparent during incident response. Rubrik's research found that the percentage of IT and security leaders who expect recovery from a cyber incident to take a day or longer has more than doubled — from 18% in 2024 to 38% in 2026 — with one-third expecting incidents involving AI agents to require even longer recovery times. Nearly nine in ten leaders want the ability to reverse an agent's actions without restoring entire systems. We've heard horror stories about AIs deleting databases while fulfilling their instructions. There should be an easier way to reverse such mistakes. "The ability to roll back actions would allow organizations to reverse specific agent-driven changes without needing a full system reset, which is critical for minimizing downtime," Mariappan explains. "What if there was an 'undo' button?" How properly architected recovery options would increase AI-agent efficiency The solution is not more governance after deployment, but better architecture before deployment. "Recovery architecture must be proactively scoped as a mandatory deployment requirement for any agentic initiative," argues Mariappan. "Organizations should mandate that any AI agent architecture includes a 'kill switch' for immediate containment and a comprehensive audit trail, ensuring that human oversight is baked into the agent's operating model from day one." That means defining acceptable-use policies, blast-radius boundaries, continuous monitoring, preconfigured rollback procedures, comprehensive audit trails and immediate kill-switch capabilities before agents begin operating. Just as importantly, recovery planning cannot remain solely the responsibility of security teams. Because AI agents increasingly affect finance, legal, operations and customer-facing processes, business continuity stakeholders should participate in deployment decisions from the beginning. "We need to stop treating operational resilience as the responsibility of any one department," Mariappan says. "Treating recovery as an afterthought leaves organizations at the mercy of fast-moving, autonomous agents that may deliberately cause disruption or simply make mistakes based on misaligned priorities." Operational resilience is just as important as the pace of innovation. Organizations that can deploy AI agents rapidly and recover from their mistakes just as quickly will define the next phase of responsible agentic transformation. "Building in recovery at the outset is the only way to ensure that operational resilience can keep pace with the velocity at which agents have the potential to cause incidents, whether malicious or accidental," Mariappan says. An In-Depth Guide to AI Get essential knowledge and practical strategies to use AI to better your security program. Learn More Paul Wagenseil Paul Wagenseil is a custom content strategist for CyberRisk Alliance, leading creation of content developed from CRA research and aligned to the most critical topics of interest for the cybersecurity community. He previously held editor roles focused on the security market at Tom’s Guide, Laptop Magazine, TechNewsDaily.com and SecurityNewsDaily.com. Related AI/ML Balancing lead generation, growth, and security in the age of AI search Solomon Adote July 27, 2026 AI search visibility demands smarter bot security, not open access. AI/ML Phishing the agent: Why identity controls are essential to secure and manage AIs Paul Wagenseil July 27, 2026 Guardrails cannot serve as the primary security boundary for AI agents. Identity controls impose firmer limits. AI benefits/risks How to Secure AI Applications in Production SC Media Editorial Intelligence, reviewed by Ankit Bhutani July 26, 2026 AI application security programs collapse at three predictable control points: system prompt integrity, output handling and runtime visibility gaps Get daily email updates SC Media's daily must-read of the most current and pressing daily news Business Email By clicking the Subscribe button below, you agree to SC Media Terms of Use and Privacy Policy . Subscribe Related Terms Banner British Standard 7799 Chain of Custody Client Common Gateway Interface (CGI) Competitive Intelligence Cookie DLL Injection Data Custodian Due Diligence You can skip this ad in 5 seconds