---
title: Organizations Face Agentic Denial of Service Risks With AI Agents
url: https://www.elseif.net/organizations-face-agentic-denial-of-service-risks-with-ai-agents
published: 2026-11-03T15:03:59+00:00
language: en
section: Agents
source: https://www.cisoadvisor.com.br/agentic-denial-of-service-nova-frente-de-dos/
organizations: Infoblox
publisher: elseif
---

# Organizations Face Agentic Denial of Service Risks With AI Agents

Organizations using artificial intelligence agents to automate cloud, IT, and security operations are facing a risk called Agentic Denial of Service, or ADoS.

Krupa Srivatsan, the senior director of product marketing at Infoblox, described this concept in a company blog post. She stated that this flaw does not crash firewalls or overload servers, but instead turns the victim's own defense automation into a weapon. ADoS prevents the effective use of AI agents, the services they protect, or the infrastructure they depend on by overloading systems with excessive calculations, actions, or requests.

One form is security denial, where the defender becomes the target. In an AI supported security operations center, alerts trigger expensive workflows. Attackers can generate enough interesting events to consume analyst attention, enrichment capacity, API quotas, and tokens. This makes defenses distracted, expensive, or slow, which helps hide a real invasion. Attackers may also poison knowledge sources to make AI assistants produce confident but incorrect summaries.

Another form is infrastructure denial, where agent activity causes the disruption. Agents rely on network paths, cloud resources, APIs, and DNS. Latency, limited APIs, and increased cloud spending can occur if a site reliability agent gets stuck in a correction loop or if a malicious input causes an orchestrator to trigger dozens of specialized agents.

A third form occurs when defenders grant agents too much autonomy to match attacker speed. A support agent might delay recovery by resetting thousands of passwords because it fails to see a provider outage. Attackers can deliberately induce this confusion through prompt injection, making the defender's response the attacker's payload.

Infoblox recommends four principles to mitigate these risks. These include blocking malicious destinations in DNS queries early where costs are low, observing behavior instead of just events, basing automation on trusted data, and limiting agent autonomy with rate limits, budgets, a kill switch, and approvals for high impact actions. The company noted that while DNS is the first layer for almost every connection, these measures do not replace governance in identity and AI orchestration layers.