{
  "id": 15930137,
  "title": "Rogue AI: When agents stop advising and start acting",
  "url": "https://urgent.news/2026/10/21/rogue-ai-when-agents-stop-advising-and-start-acting",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-21T18:22:39.000Z",
  "source": {
    "name": "Computerworld",
    "slug": "computerworld",
    "url": "https://www.computerworld.com/article/4238623/rogue-ai-when-agents-stop-advising-and-start-acting.html"
  },
  "original_language": "en",
  "account": "Since the advent of AI technology, most commercial applications have focused on providing advice and suggestions. However, a new generation of agentic AI systems are emerging that can autonomously plan and execute complex tasks, from querying databases to configuring systems and even closing tickets without human intervention. A recent Ivanti survey found that 87% of cybersecurity professionals believe integrating agentic AI is a top priority, and 77% feel comfortable allowing autonomous systems to act without human review. While the allure for IT teams overwhelmed by constant demands is clear, the risks are equally significant. A rogue AI that provides a faulty response may inconvenience users, but one that carries out a harmful action could cause widespread system failures. The term \"rogue AI\" often conjures images of malevolent machines, but in reality, the danger lies in more mundane oversight failures. An AI tasked with freeing up disk space may inadvertently delete critical log files, while a configuration drift corrector could apply changes to unintended devices. These issues arise when agents possess more permissions than necessary, lack sufficient context, or operate without any checkpoints before performing high-impact actions. Traditional governance frameworks designed for human-generated outputs have failed to keep pace with the new challenges posed by agentic AI. While output governance remains important - ensuring responses are accurate and sensitive data is not exposed - effective governance now extends to the actions agents can take. Teams must define the precise permissions agents require for each task, establish conditional approvals based on action risk levels, and implement safeguards that limit potential blast radii and allow for reversibility. Assigning human owners to each agent, documenting the AI's purpose and scope, and maintaining comprehensive audit trails are essential to establishing accountability. As AI agents operate with increasing autonomy, compliance becomes a critical concern. Unauthorized actions by an AI could expose organizations to significant regulatory liabilities. Governance must be baked into the AI's lifecycle from the outset, considering the permissions granted, the environments in which it operates, and the approval processes in place. Rather than micromanaging every single action, an effective governance model reserves human oversight for high-risk, unusual, or irreversible decisions. As agentic AI capabilities become integrated into the tools already used by IT teams, organizations must conduct thorough inventories to identify where autonomous actions are possible and assess the associated permissions. By establishing clear action governance policies, teams can responsibly harness the benefits of autonomous AI while mitigating the associated risks. The future of IT operations lies in striking the right balance between automation and human oversight, ensuring that the benefits of AI are realized without sacrificing security, compliance, or accountability.",
  "summary": "For the past few years, most enterprise AI has been advisory. It summarized tickets, drafted knowledge articles, and suggested next steps. A person still decided what to do. That’s changing. Agentic AI systems can plan multistep tasks and carry them out: querying systems, triggering workflows, changing configurations, and closing tickets with little or no human involvement. Ivanti’s 2026 State of…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}