The second layer of defense for AI agents.
Your agents are already taking consequential actions. Proving they stayed inside the rules is harder than it should be.
Verya checks every consequential action your agents take against your rules and the regulations that apply, holds what breaks them, and keeps independent evidence of it all.
Why Verya
Agents decide and act. Proving they stayed inside the rules is the hard part.
Approving a payment, granting access, releasing a record, filing a submission: the agent decides and the action happens. The controls that exist live inside the same system that took the action, and the logs describe what happened rather than prevent it.
Some teams have guardrails in place and some are only starting. Either way, what is hard to produce is independent evidence, on request, that a specific action was checked against a specific rule before it happened. That is the gap an auditor or your own board will probe. Verya closes it.
Demo
See Verya in action.
A walkthrough of the agent overview, the review queue and the regulation view.
How it works
From connected to controlled in four steps.
- ConnectAdd a small SDK check around an existing function, or one webhook step. Verya starts in shadow mode, so nothing is blocked.
- MapVerya reads each agent against your own policy and the regulations that apply, and shows where exposure sits.
- ControlImport the guardrails you already have, or start from none, then create new controls as monitoring shows what your agents do. Each hold waits for a named person.
- ProveEvery decision is recorded append-only, so you can show what was checked, when, and who decided.
Visibility
Every agent in one place.
Register an agent and it appears on one page, grouped by what it does. See its mode, what has been checked, flagged or stopped, and the next step waiting on you.
Posture
Know where every agent stands.
Policies, controls, regulations, assessments, operational risk and core info, each shown as a plain ratio of what is done. Click any axis to go and do the work.
Regulation
Every requirement mapped to a control.
See which laws apply to which agents, and whether a control covers each requirement. Verya proposes every mapping and two qualified people confirm it, so gaps are visible, never assumed away.
Control
Hold risky actions. A named person decides.
Bring your existing guardrails or start without any, and add new controls as Verya shows how your agents actually behave. Turn them on one at a time. Actions that break a rule are held for review, grouped by cause so one fix clears every action it affects. Each decision records who made it.
The regulatory picture
Rules for AI agents are taking shape in every major market.
Boards are already asking how the firm knows its agents are under control. Regulation is moving from principles to expectations that have to be evidenced.
Rules are becoming law
The EU AI Act is in force, and other jurisdictions are following with their own rules on how AI is governed and overseen.
Regulators expect accountability
Financial regulators, including the UK’s FCA, expect firms to stay accountable for what AI does on their behalf, however it is built or bought.
Standards set the bar
Frameworks such as ISO/IEC 42001 and the NIST AI Risk Management Framework are becoming the common language for showing AI is under control.
When an agent gets it wrong, the firm and its senior managers answer for it, not the vendor.
Verya maps each agent against the regulations that apply today and tracks new ones as they come into scope, so your board can see where exposure sits and what evidence exists.
Who it is for
One layer, three teams who need it.
Risk and compliance
See regulatory and operational exposure mapped, explained and ready to share.
Engineering
Add a few lines to agents you already run. Works across Claude, GPT and Gemini.
Audit and the board
Produce independent evidence, on request, that a specific action was checked against a specific rule.
Independent by design
Evidence only counts if it does not come from the system being checked.
Verya sits outside your agents and their vendors, so the evidence it produces is independent of the systems being checked.
- Outside your agentsVerya checks actions independently of the agent and its vendor, so you are not relying on a system to mark its own work.
- Any AI providerWorks across Claude, GPT and Gemini today, and is built to plug into whatever comes next.
- Shadow mode firstSee what Verya would catch before you switch anything on.
About
Built by people who have been asked to prove it.
Verya is built by people who have worked in risk, compliance and partnerships at Stripe and some of the world’s largest financial institutions, and who have seen how hard it is to prove that controls actually worked. We built the tool we wanted when an auditor or a regulator asked.