Mitiga and Modern Industrials bet on more regulated AI agents

Mitiga launches a platform to detect AI agent abuses from cloud, SaaS, and identity logs. Meanwhile, Modern Industrials raises $6 million to deploy custom agents in industry.

On November 19, 2026, two announcements illustrated the same trend: AI agents are moving from demos to real systems, but with more safeguards. Mitiga launched a security platform dedicated to AI agents, while Modern Industrials raised $6 million to automate tasks for manufacturers and distributors. In both cases, the challenge is the same: let agents act without losing control over what they do or access.

Des agents IA plus utiles, mais aussi plus traçables

Mitiga, a cybersecurity company, offers AI Agent Security, an agentless platform that detects abusive agent usage based on behavioral baselines, i.e., normal activity profiles established from logs and permissions. The tool does not block an action before it happens; it observes, correlates, and reconstructs what the agent did, then relies on the client’s existing security integrations to contain the incident. Ofer Maor, co-founder and CTO, summarizes the problem: “An agent doesn’t need to break in when it already holds the credential. To the audit log, it looks like a valid login.”

This approach addresses a well-known blind spot for companies: an agent can act via service accounts, OAuth tokens, API tokens, or delegated sessions without immediately triggering classic alerts. Mitiga says it covers in-house agents built on Bedrock or Azure AI Foundry, agents provided by vendors via OAuth, as well as code assistants, personal AI tools, and unreviewed MCP servers. The product relies on already available logs and APIs, without SDK, proxy, or agent code modification. Convenient to avoid technical detours; less so if the company lacks exploitable logs.

La visibilité devient la vraie ligne de défense

The product documentation also sets its limits. Coverage depends on systems that actually expose agent activity in cloud, SaaS, identity, and AI logs. Without this telemetry, the platform mainly detects downstream behaviors. Prompt-level signals are visible only when supported platforms expose them, notably via Bedrock, Azure OpenAI, Agentforce, or Copilot logs. Mitiga also indicates possible responses such as token revocation, identity quarantine, or rollback of a change via client integrations.

The company’s promise is based on a simple idea: securing an agent is not just about monitoring its intentions but tracking its traces. Mitiga also says it normalizes telemetry from over 50 platforms in its Cloud Security Data Lake and retains data for more than 1,000 days. The platform claims 95% automated security operations center workflows, 100% triaged alerts with auditable verdicts, and under three minutes between event collection and actionable response, according to its internal figures.

Dans l’industrie, l’automatisation progresse à pas mesurés

Modern Industrials, for its part, takes a less defensive but equally cautious path: building AI agents for industrial manufacturers and distributors without replacing existing software all at once. The company announced a $6 million seed round led by Glasswing Ventures, with TMV co-leading and participation from a16z speedrun and other investors. Austin Mao, general manager, wants to use this funding to accelerate development, grow the team, and further deploy its approach centered on existing business processes.

Its agents target still very manual tasks stuck between emails, spreadsheets, and management systems. Use cases cited include quotes, order taking, demand planning, supplier management, inventory, operations, and customer support. The company works with already installed ERP systems, including Oracle NetSuite, Microsoft Dynamics 365, JD Edwards, Acumatica, Epicor Eclipse, Epicor Prophet 21, and Infor. The idea is not to start from scratch but to graft agent workflows onto the existing setup.

Des droits d’écriture accordés avec parcimonie

The company emphasizes gradual deployment. Its systems remain read-only until the client authorizes write permissions, which are limited to specific actions. Teams can first have everything reviewed by a human, then expand automation if the system proves reliable on real workflows. This graduated permission logic spans sales, purchasing, finance, and operations functions. It avoids a big leap into the unknown, which in industry is generally a good idea.

The funding round also relies on customer testimonials. Steve Rask, CEO of AC Summit and MJB Wood Group, believes the technology “will save hundreds of hours and millions of dollars.” John Chisholm, president of Atlantic Forest Products, says the team has solved every new problem presented. On the investor side, Rudina Seseri, founder and managing partner of Glasswing Ventures, sees reasoning and agentic AI as a way to create gains hard to achieve with traditional software. Soraya Darabi, co-founder and managing partner of TMV, highlights the founders’ method: understand each client, then build around their business.

Une même question revient: qui contrôle quoi, et jusqu’où ?

These two announcements do not address the same market, but they tell the same story. On one side, enterprise security must learn to monitor systems acting with legitimate credentials. On the other, industrial companies want to automate more without breaking their processes or ERPs. In both cases, value no longer comes only from AI that “does,” but from AI that can be connected, limited, audited, and, if needed, taken back under control. It’s less flashy than a slogan but much more useful for an IT director or industrial operator.

By the numbers

  • $6 million — seed raised by Modern Industrials on November 19, 2026.
  • Over 50 platforms — telemetry normalized by Mitiga’s Cloud Security Data Lake.
  • Over 1,000 days — data retention period announced by Mitiga.
  • 95% — automated SOC workflows claimed by Mitiga.
  • Under 3 minutes — time reported between collection, triage, and response at Mitiga.

To read