Mitiga Launches AI Agent Security to Stop Rogue or Malicious Agents at Machine Speed

PR Newswire

The platform sees every agent where it runs, controls what it can reach, and stops and contains it in runtime when it turns.

NEW YORK, Nov. 19, 2026 /PRNewswire/ — Mitiga, which delivers Agentic Runtime Security for the modern infrastructure, today launched AI Agent Security, extending its runtime detection, control, and Zero-Impact Breach Prevention to the AI agents already working inside cloud, SaaS, and identity systems on credentials the enterprise issued. AI Agent Security sees and controls every agent, stops AI attacks, and contains breaches at machine speed.

Mitiga Logo

As AI agents access sensitive data, interact with business applications and execute tasks through service accounts, OAuth grants, API tokens and other identities, traditional security infrastructure can’t keep pace. Built to monitor human users and endpoints, traditional tools have little insight into what autonomous agents are doing, leaving organizations exposed to compromised agents and unauthorized activity carried out through legitimate credentials.

Mitiga AI Agent Security discovers agents across enterprise environments, monitors their behavior against their permissions in runtime, and catches the moment legitimate access turns malicious. Mitiga correlates activity across those environments, so security teams can investigate the threat, reconstruct what the agent did, and stop and contain it with reversible actions before the attack causes business impact. 

“AI agents are the newest privileged users in the enterprise, and most security teams can’t say which ones can reach critical data, let alone what those agents did last week,” said Ofer Maor, Co-Founder and CTO at Mitiga. “An agent doesn’t need to break in when it already holds the credential. To the audit log, it looks like a valid login. Mitiga reads the behavior behind it across cloud, SaaS, and identity, and when an agent steps outside its purpose, we stop it and contain it in runtime.”

Three Kinds of Agents, One Blind Spot
Engineering builds homegrown agents on Bedrock, Azure AI Foundry, or a custom stack. Business units connect vendor agents such as Salesforce Agentforce, ChatGPT, and Anthropic through an OAuth grant nobody revisits after the approval meeting. Employees add their own, from coding assistants to personal AI tools on work identities to MCP servers and skills no one reviewed. Most security teams can only see one of the them.

Guardrails shape what an agent may ask for. Posture tools map what it could reach. A stolen token, a vendor’s OAuth grant, or an agent nobody registered passes through neither. Mitiga cross-references static configuration with live runtime activity, so privilege drift and unauthorized access surface as they happen and well before the next scan.

Four Scenarios Mitiga AI Agent Security Covers
Third-party agent compromise. A sanctioned vendor agent starts reading in bulk off-hours and picks up new scopes in two SaaS apps. Mitiga ties the grant, the API calls, the data reached, and the downstream cloud activity into one incident, then revokes the grant.

  • Homegrown agent overreach. An agent built on Bedrock or Azure AI Foundry inherits a broad role and drifts into resources outside its task. Mitiga shows the gap between what the role allows and what the agent does, and flags the drift before it becomes an incident.
  • Coding agent and MCP supply chain. A developer installs a skill or MCP server carrying hidden instructions or a credential-exfiltration path. Skillgate, Mitiga’s free scanner, catches it before the agent acts, and if the agent already ran, Mitiga shows what the developer’s credentials did next.
  • Cascading agent actions. One compromised agent triggers tools, services, and other agents at machine speed. Mitiga correlates the sequence into one attack timeline and identifies the safest point to contain it.

How It Works
Mitiga establishes behavioral baselines for agents across everything they touch, monitoring the systems they access, actions they perform, and identities they use. It also connects available prompt-level signals with downstream activity to identify when malicious instructions influence agent behavior. Through AI-assisted investigation, the platform correlates cloud, SaaS, identity, and AI activity into a unified incident record. Mitiga automates 95% of SOC workflows and triages 100% of alerts, providing an auditable verdict for each.

When an agent turns, Mitiga can revoke its token or session, quarantine its identity, and reverse the change, with each action scoped to that agent and the smallest possible blast radius. Actions run through integrations already in the customer’s stack, and what can be reversed depends on what each integration supports.

Nothing is installed in the agent’s path. Mitiga is agentless, working from the logs and APIs a customer’s cloud, SaaS, identity, and AI platforms already produce, with no gateway, SDK, proxy, or code change inside any agent. The same logs reach back as far as retention does, so teams can investigate agent activity that predates any agent control. Mitiga detects in seconds and responds at machine speed – moving from event to triage, to actionable response in under 3 minutes.

Panoramic Visibility Across Enterprise Environments
AI Agent Security runs on Mitiga’s Cloud Security Data Lake, which normalizes cloud, SaaS, identity, and AI telemetry from 50+ platforms and retains it for 1,000+ days. Context Engineering turns that telemetry into investigation-grade context before detection runs, and 5,000+ continuously tuned detections work from it.

Activity Flow shows which users and non-human identities acted, through which AI platforms and connectors, and which applications each one reached. Because the evidence is the logs a customer already collects, a question like “what did our agents do in the last 30 days?” has an answer on day one.

The morning after an agent misbehaves, security leaders ask what it did, how far it reached, and what to do right now. Mitiga answers from one record. That’s Zero-Impact Breach Prevention applied to agents instead of endpoints, delivered through response actions that stay scoped and reversible.

Availability
Mitiga AI Agent Security is available now. Organizations can learn more at https://www.mitiga.io/ai-agent-security and request a demo at https://www.mitiga.io/demo.

About Mitiga
Mitiga delivers Zero-Impact Breach Prevention for AI agents. Its AI Agent Security platform finds every agent acting across the modern infrastructure, baselines each agent’s behavior across everything it touches, and reconstructs each agent’s activity through the identities it uses. Mitiga detects rogue agents and AI attacks in runtime, then stops and contains them before they impact the business. Agentless and easy to deploy, AI Agent Security runs on a normalized, contextualized Cloud Security Data Lake that holds investigation-grade data across cloud, SaaS, identity, and AI, with 1,000+ days of history and 6,000+ continuously tuned detections. Teams know what an agent did, how far it reached, and what to do right now. Zero-Impact Breach Prevention, for every agent.

Let them come. 

Media Contact
5WPR — mitiga@5wpr.com — 212-999-5585

Cision View original content to download multimedia:https://www.prnewswire.com/news-releases/mitiga-launches-ai-agent-security-to-stop-rogue-or-malicious-agents-at-machine-speed-302944640.html

SOURCE Mitiga

About The Author