← Back to blog

Your people are already using AI. Make being compliant the easy path.

By Michael Cooper · Founder

Compliance

They are not dodging the rules — they are routing around a compliant path that does not exist yet. Pave one they can actually take, and the documentation takes care of itself.

Ask a CTO whether they know which AI tools and agents are touching their internal systems, and most will tell you they have it handled. The data says otherwise — and the gap is not a people problem, it is a path problem.

In Okta's AI Agents at Work 2026 survey, 90% of executives said they were confident in their visibility into AI tools — while 52% of knowledge workers admitted to using unsanctioned ones. The blind spot is structural. Salt Security's 1H 2026 State of AI and API Security report found 48.9% of organizations are entirely blind to machine-to-machine traffic, and 48.3% cannot tell a legitimate AI agent from a malicious bot. The average enterprise is now running dozens of agents it did not centrally deploy — and 88% report a suspected or confirmed AI-agent security incident.

It is tempting to read those numbers as a discipline problem. They are not. The people wiring agents into CRMs, internal APIs, and the ERP are capable individual contributors in finance, operations, and engineering, trying to get work done faster. They are not dodging the rules. They are routing around a compliant path that does not exist yet.

People route around friction, not rules

Here is the thing every platform team eventually learns: if the sanctioned path is slow, unclear, or missing, people do not stop — they go around it. Shadow AI is not rebellion. It is water finding the downhill route.

Which means the fix is not a crackdown. It is a better path. Make the compliant way to use AI the easiest way to use it, and people will take it — not because you forced them, but because it is the path of least resistance and it does not slow them down. The moment “do it compliantly” costs more effort than “just call the API,” you have lost. The moment it costs less, compliance stops being a fight.

Pave the path — and let the agent walk it

That is the job AGLedger is built for. Because we speak plain API and HTTP, you point your existing gates and tools at us; there is no new control plane to stand up. When someone connects a new agent, it is directed to a happy path it can navigate on its own — declare what it intends to do, do the work, and the documentation is produced as a byproduct. No human fills out a form. No one files a ticket. The easy path is the compliant path.

And you move at your own pace, in two steps. First, ask your agents to be compliant — point them at the easy path and let them declare what they intend to do. Most will, because it costs them nothing and slows them down not at all; nothing is gated, work flows as it does today, and you simply start getting a signed record of what your agents intend and do. Then, when you are ready, block the ones that are not — flip from asking to requiring, and undocumented work stops at the door. Invitation first, enforcement when it suits you, and you never halt the business to get there. That is the whole trick: compliance that does not feel like a tax gets used.

One boundary worth being precise about: AGLedger is a notary at a boundary, not a discovery scanner. We do not crawl your network to find shadow agents — that is what CASB, non-human-identity governance, and API-security tools are for. We make the sanctioned path the one worth taking, and we sign what crosses it.

The record writes itself — and it is more than a log

This is not a rip-and-replace of your SIEM, your observability stack, or your ERP. It sits over them and makes them more useful.

Logs are forensic: they reconstruct what happenedafter the fact, from effects. They can tell you “user Ted was added to the database.” What they structurally cannot do is record intent before the action — and an agent's own after-the-fact account of what it did cannot be trusted, because it can hallucinate, forget, or rationalize. AGLedger captures a signed statement of intent and success criteria at the moment before work begins, Ed25519-signed and hash-chained, so the result can be measured against a commitment the agent cannot fake or forget.

Your log still says Ted was added. AGLedger tells you the whole story behind it — that an agent intended to add Ted, who the accountable principal was, and what was supposed to be true when it finished. That is worth something even for a single agent. It becomes decisive the moment work is delegated across agents, teams, or companies, where logs can no longer stitch the chain together.

And the account is audit-ready whenever you are

Here is the quiet payoff. Every action that travels the easy path accrues into an append-only, tamper-evident record — exactly the kind of account regulators are going to ask for.

The EU AI Act's record-keeping and logging obligations (Articles 12 and 19) survived the 2026 Digital Omnibus simplification untouched; only the high-risk deadlines moved out — to December 2027 for standalone Annex III systems. That is not a reason to wait. It is runway to build the account properly, as a byproduct of how your teams already work, instead of reconstructing it under audit pressure later.

To be precise about what we provide: AGLedger is the evidence pattern, not a compliance certificate. Your program still owns the policy, the methodology, and the decisions. But because the documentation has been accruing the whole time your people took the easy path, it is ready for EU AI Act record-keeping whenever you are. See the EU AI Act capability crosswalk for how it lines up with Articles 12, 19, and the rest.

The Developer Edition is free and fully functional — not a trial, and it runs in production. Install it, ask your agents to start declaring what they do, and watch the record build; block the holdouts whenever you are ready. When you want to run it on your own external database at scale, that is Enterprise — the paid trigger is the database, never the feature.

Sources & further reading