Filip Perica

Building trust in AI

🤖 Director of BD @nuggetsPAYandID
🏉 International rugby player

[Filip Perica @Filip_Perica]

@KhanAbbas201 Garmin

[Filip Perica @Filip_Perica]

book here: https://t.co/OKOE9eAbBE

[Filip Perica @Filip_Perica]

Hey everyone 👋, is your company deploying AI agents? 🤖

Can you prove what your AI agents are authorised to do at the exact moment they take an action?

If the answer is NO, then this is for you. 👇

We're taking on a select group of founding design partners for our Authority Control Plane (ACP).

In a FREE 30-minute Agent Authority Exposure Review, our engineering team will review your current agent architecture, identify authority gaps, and demonstrate what implementing verifiable agent authority could look like.

Link in the comments and happy to answer any questions.

[Nuggets ™ @nuggetsPAYandID]

We've launched the Authority Control Plane.

One place to define what your AI agents can do, enforce it the moment they act, and prove every action.

No authority, no action. Trust that travels.

https://t.co/yDNOajJaIX

[Filip Perica @Filip_Perica]

This 👇

[Seema Khinda Johnson @KhindaSeema]

I would like to introduce the Authority Control Plane. It decides what an agent is allowed to do, proves it, and lets you ship your agents.
Thank me later ❤️
https://t.co/vt7Ar0xG1S

[Filip Perica @Filip_Perica]

If you're building agents with LangChain or LangGraph, check this out.

No more hoping your agent respects scope. langchain-nuggets intercepts every tool call before execution, checks delegation scope and invocation limits and emits a signed proof artifact on every allowed call.

Just drop it into your existing agent. No account required to run the cross-org demo locally.

Install
pip install langchain-nuggets
npm install @ nuggetslife/langchain-nuggets

https://t.co/czwNGBgM4f

[Seema Khinda Johnson @KhindaSeema]

Access decides what systems your agent can reach. Authority decides whether this specific action, for this user, right now, is one it's allowed to take.

We just shipped langchain-nuggets, for LangChain and LangGraph developers. It enforces that at execution time and leaves a proof anyone can verify, even across orgs and clouds.

Trust travels with the agent. Every allowed action carries a receipt a partner or a regulator on a different cloud can verify for themselves, without access to your systems, without trusting your logs, and without exposing your raw data. Hashes, not values.

See it running: the cross-org demo, locally, with one ALLOW, four DENY paths, intent binding, and proof verification. https://t.co/VZonCuUqCJ

Open source. Python and TypeScript.
pip install langchain-nuggets
npm install @nuggetslife/langchain-nuggets
Code: https://t.co/lGO3iIOZtk
Full write-up: https://t.co/yFEyWyNvmx

[Filip Perica @Filip_Perica]

TechRound has named us one of the 45 most exciting AI companies in the world.

When we built our AI governance framework, we knew we were probably a bit ahead of the curve, but the market caught up fast and what felt early then feels urgent now.

AI agents are currently acting across clouds and organisations with no verifiable authority This is why we govern AI at the point of execution, so every action is bound to a verified identity, a confirmed authority and a tamper-proof audit trail.

More to come.

[Nuggets ™ @nuggetsPAYandID]

Proud to be named in @techrounduk's AI45 2026, celebrating the most exciting AI companies across the world!

A big thanks to the judges and congrats to everyone on the list. Full list 👇

https://t.co/lAigdddwmS

[Seema Khinda Johnson @KhindaSeema]

https://t.co/Z6dFNl8Xbw

[Seema Khinda Johnson @KhindaSeema]

FIS and Anthropic launched their financial crime AI agent on 4 May, with BMO and Amalgamated.
If your AI agent takes an action, can you prove it was authorised? 18 questions to test your deployments before production. https://t.co/0inazTJu2a

[Filip Perica @Filip_Perica]

Here you go 👉 https://t.co/75ibKg8jPX

[Filip Perica @Filip_Perica]

FIS and Anthropic launched their financial crime AI agent earlier this month, with BMO and Amalgamated as the first banks using it.

lately, i'm probably speaking more with compliance than product. I feel like the same question keeps coming up.

If an AI agent takes an action inside the bank, can we prove it was authorised? That specific action, by the right person, within the limits it was supposed to operate in.

most governance frameworks today were built around access, the evidence usually stops at the access log.

Article 26 of the EU AI Act starts applying to high-risk AI systems from 2 August 2026 and asks for evidence most frameworks don't really produce today.

the FCA is heading in a similar direction under SM&CR.

we've published 18 questions for compliance and risk teams to go through before these systems go live.

Free, link in comments.

What's the hardest question for you to answer?

[Seema Khinda Johnson @KhindaSeema]

https://t.co/rQmErLQbUF

[Filip Perica @Filip_Perica]

if you're thinking about adding AI system and your vendor cant answer the questions below 👇, thats a red flag.

You’re probably taking on more risk than you think.

Luckily, if you’re already integrated and see the gaps, @nuggetsPAYandID can help close them.

[Filip Perica @Filip_Perica]

most companies deploying AI agents today cannot answer three questions that regulators, auditors and boards will increasingly ask.

Who authorised this action? On whose behalf was it taken? Under what authority?

Great read 👇

[Seema Khinda Johnson @KhindaSeema]

https://t.co/9gjeFAFBYq

[Filip Perica @Filip_Perica]

Dm for me if you are interested in enabling a trust layer for your agents.

[Nuggets ™ @nuggetsPAYandID]

Most companies can control who gets into their AI systems.
Almost none can prove what those systems did once they were in.
@BiometricUpdate covered our Enterprise AI Governance Framework & why the control point for AI has shifted from access to execution.
https://t.co/WFTctI1vDv

[Filip Perica @Filip_Perica]

no trust, no party 💁

[Filip Perica @Filip_Perica]

@KhanAbbas201 Not surprised

[Filip Perica @Filip_Perica]

AI is getting action,
you might want protection.