Automation Money Is Pouring In and Nobody Checked the Locks

Plumbing is invisible until it leaks. You never think about the pipes behind your walls, and then one morning your ceiling is on the floor. That is roughly where the automation tooling market sits right now: everyone is admiring the fixtures, nobody is inspecting the joints.

The news that prompted this: n8n, the Berlin-based no-code automation company, raised $180 million at a $2.5 billion valuation, with Nvidia among its backers. That is a serious number for a company whose product is, at its core, a way to wire systems together without writing much code.

I have no complaints about the raise. From where I sit as a security researcher, though, the number tells me something more interesting than “investors like automation.” It tells me that the connective tissue between AI models and business systems is about to get a lot more capital, a lot more users, and a lot more attack surface — probably in that order.

Why Automation Platforms Are Different Targets

Most security thinking about AI has fixated on models. Prompt injection, jailbreaks, training data poisoning. Those matter. But a model on its own is a very expensive text generator. It becomes consequential when it can do things: read your CRM, send email, move money, open tickets, write to a database.

Automation platforms are exactly the layer that grants those abilities. A workflow builder is, functionally, a credential vault attached to an execution engine. It holds API keys for a dozen services because that is the entire point. It runs on triggers so it does not need a human present. And increasingly it h

Stack those three properties and you get something worth thinking carefully about:

• Concentrated credentials. One compromised automation account can mean access to every service that account was configured to touch.
• Unattended execution. Scheduled and event-driven workflows fire without anyone watching. Anomalous behavior can run for a long time before someone notices.
• Non-deterministic logic. When a model decides which branch to take, the same input does not always produce the same output. Traditional testing assumes it does.

No-Code Means No Security Review

The selling point of no-code is that non-engineers can build things. That is genuinely useful. It is also a governance problem that most organizations have not confronted.

When a developer ships code, it passes through review, static analysis, a CI pipeline, maybe a security sign-off. When someone in operations builds a workflow that pulls customer records, sends them to a model for summarization, and posts the result into a chat channel, none of that happens. There is no pull request. There is no reviewer. The person who built it may not know that customer data just crossed an organizational boundary.

This is not a criticism of no-code tools. It is an observation that the tooling for building these systems has raced ahead of the tooling for auditing them. Security teams are being asked to secure things they cannot see.

The Questions Worth Asking

If your organization uses automation platforms — and if you are a company of any size, someone in your organization does, whether you know it or not — a few practical starting points:

• Can you produce a list of every active workflow and what credentials each one holds?
• Who has permission to create workflows, and does that permission distinguish between reading data and writing to external services?
• Are workflow executions logged somewhere your security team actually monitors?
• When a model output feeds into an action, is there any validation between the two, or does the model’s word become an API call directly?
• What happens when an employee with workflow access leaves?

That last one catches people out more often than you would expect. Automations outlive their authors.

Funding as a Forecast

Money arriving at this layer is a reasonable signal about where systems are heading. Nvidia backing an automation company rather than only model companies suggests a view that the value is in the wiring, not just the intelligence. I think that view is correct. I also think it means the security conversation needs to move with it.

Defenders have spent the last couple of years learning to think about model inputs. The next problem is model outputs with hands attached — and the platforms that attach them are now very well funded and growing fast.

Better to inspect the pipes now than to explain the ceiling later.

🕒 Published: September 9, 2026