Verdict

The Inspiration

The idea for Verdict was born out of a pretty terrifying realization: we’re handing autonomous AI agents the keys to the kingdom, but we forgot to install any locks on the doors.

As developers, we were watching teams build these incredible AI agents to handle sales, support, and finance. But there was a glaring issue. These agents are given "Master Keys"—direct API access to Salesforce, Gmail, and highly sensitive internal databases. The problem? An AI doesn't have 20 years of social training, corporate etiquette, or a "gut feeling" about what might get the company sued.

If you tell a finance agent to "optimize cash flow," it won't hesitate to bypass a mandatory two-person approval process or accidentally drop sensitive customer data into an unsanctioned spreadsheet if it thinks it’s finding a faster shortcut. We realized that if the "Agentic Era" is actually going to work in the real world, AI doesn't just need to be smarter. It needs a conscience.

So, we built Verdict to be that conscience.

What It Does

At its core, Verdict is a developer-first authorization API. Think of it as a bouncer or a secure checkpoint standing between your AI agent and the tools it’s trying to use. Instead of crossing your fingers and hoping the AI behaves, Verdict actually enforces your business rules in real time.

Here is how it works in practice:

The Checkpoint: We made it dead simple. Developers just wrap their agent’s tool-calling functions in a single line of Verdict code. Before the agent can actually execute an action—whether that's sending a client email, moving money, or pulling database records—it has to ask us for permission first.

The Decision Engine: We evaluate what the AI is trying to do and the data it's trying to use in under 50 milliseconds. We instantly reply with one of four decisions: Allow (it's safe), Deny (it breaks a rule), Mask (hide the sensitive PII first), or Require Approval (this is too risky, ask a human).

The Control Surface: Developers love the API, but security is a team sport. We built an Admin Dashboard so the non-technical folks—like Legal, Security, and Finance—can easily translate their "unwritten company rules" into hard-coded guardrails without needing to write a single line of code.

The Black Box: We record everything. Every decision, block, and human approval is saved in an immutable audit trail. It’s essentially a flight recorder for your AI, giving you exactly what you need to prove SOC2, HIPAA, or GDPR compliance.

What's Next

Right now, we are transitioning Verdict from a high-fidelity concept into a working technical MVP. Our main goal for the immediate future is shifting AI security from being purely reactive to proactively smart:

Blink-and-You-Miss-It Latency: We’re optimizing our engine to get decision times under 10ms. Security needs to be completely invisible so it never interrupts the agent's "train of thought" or slows down your app.

The "Policy Graph" Library: We're building a library of pre-configured, compliance-hardened templates. Imagine a startup being able to make their AI "HIPAA-Safe" in under five minutes just by toggling a switch.

Learning the Company Culture: We want Verdict to learn from your team. As managers manually approve or deny flagged actions in the dashboard, Verdict will observe those decisions and suggest tweaks to your policies, eventually learning your organization's unique risk appetite.

Playing Nice with the Ecosystem: We’re building one-click integrations to push our data directly into major security tools like Splunk and Datadog, making Verdict the central nervous system for all enterprise AI forensics.

Built With

• html
• pitch