C:\>DIR Global ‘Agentic Regulator’ Hackathon 2026

Final Round & Winners

The Hackathon in numbers

The C:\>DIR Global 'Agentic Regulator' Hackathon attracted 336 high-quality team submissions from over 65 countries – spanning regulators, policymakers, start-ups, developers, tech vendors, infrastructure providers, professional services firms, academics and students.

Following a rigorous review process, the organisers selected 36 teams, on average six per problem space, for the final prototype build round.

Notably, 19 of the 36 finalist teams were from, or affiliated with, regulatory authorities, including central banks, financial regulators and other public bodies.

Infographic showing data about applicants: 942 individuals applied, with 336 team submissions, 65 countries represented, 56 from private sector, NGO, academic institutions, 48 partners, 36 teams for final prototype build, 19 finalist teams with regulatory authorities, and 6 problem spaces.

Final Build Round

Over the course of a week, teams were given access to sandboxes and synthetic datasets in NayaOne and vouch.finance. Teams had one week to develop their prototypes and guardrails.

Teams were offered mentorship, knowledge sessions on Agentic AI & Agentic AI in a Regulatory Context, alongside Pitch Clinics & Tech Clinics. 100% of the teams submitted code or working applications for assessment, a presentation & demo.

The Virtual Demos + Live Voting + C:\>DIR Fellows Judging

Over the course of a week, teams presented their Demos for judging by regulators from around the world to get to the Final 6.

'Virtual Demo + Live Voting' Session 1

15 September 2026, 12:00–15:45 BST

AI-Driven Fraud & Scams

Agentic Payments, Commerce & their Oversight

Agentic Financial & High-Stakes AI Advice

'Virtual Demo + Live Voting' Session 2

16 September 2026, 12:00–15:30 BST

Know Your Agent (KY-A), Digital Verification & Digital Public Infrastructure

Market Manipulation, Agentic Herding & Stability

Decentralised Market Infrastructure, Smart Contracts & AI Agents

Gender disaggregated data

Female gender symbol in blue on a white background

Overall, the C:\>DIR ‘Agentic Regulator’ Hackathon 2026 initially received approximately 20% of concept notes as an overall percentage of applications from women.

As the Hackathon progressed, however, the percentage of female representation grew.

The 36 teams selected for the final build round had approximately 31% female representation, which rose to 44% amongst the Final 6.

Demos + winners

The Final 6 teams pitched their solution to 15+ senior global regulatory leaders who assessed the solutions developed. The winners were announced as part of the C:\>DIR Summit which was held in Cambridge, UK on 18th September 2026.
Recording of the panel announcing the winners is available:

Coming soon

Team presentations & demos are included below for each problem space

Blue circular icon with a bar chart and upward arrow, labeled "Market Manipulation & Herding."

Create agentic solutions that will monitor and mitigate against potential market manipulation and herding by autonomous AI agents.

Team Pheonix

INDIA
Securities and Exchange Board of India (SEBI)

Uses digital twins and trading agents to test draft regulations in a synthetic securities market before they take effect.

Presentation | Demo
VeltoraCore

BRAZIL
VeltoraCore, Central Bank of Brazil, Ripple, Inter

Detects AI-driven herding and contagion in tokenized markets, then simulates targeted liquidity interventions to help authorities respond.

Presentation | Demo

COLFI

UK
COLFI, Goldman Sachs

A stress-testing lab that detects whether shared AI models amplify market stress and tests safeguards such as action limits, slower execution and model diversity.

Presentation | Demo
Festina Lente

SINGAPORE, URUGUAY
Singapore Management University, Viom, Software and Information Industry Association, AWS, Quantitative, Central Bank of Uruguay

An exchange-level system that slows trading when AI-driven herding emerges, giving supervisors time to assess the risk and decide whether to intervene.

Presentation | Demo
Hickory AI

SWEDEN, SWITZERLAND
Sveriges Riksbank, Expedia Group

Tests whether private trading models respond similarly to a shared stress scenario, revealing risks of market herding without exposing the models.

Presentation | Demo
No Human Intelligence

UK
LSE, Oxford, Impact Advantage

A synthetic-market sandbox testing whether surveillance tools and circuit breakers can detect and contain AI-driven herding before regulators rely on them.

Presentation | Demo

Icon of a document with circuit lines and the text "Decentralized Infrastructure & Agents"

Wallet-connected AI agents now can transact directly with self-executing smart contracts - routing orders, managing collateral and rebalancing liquidity across DeFi and tokenised markets. What agentic solutions can help public authorities to understand and monitor agentic on-chain activities and mitigate potential systemic vulnerabilities?

SUPPORTED BY THE ETHEREUM FOUNDATION

YU-SAM POWER

UK
University College London

An agentic framework for assessing the risk & identifying smart contract attacks for consistent enforcement and international collaboration.

Presentation | Demo
AMW

UK, INDONESIA
Cambridge Judge Business School, Brawijaya University

Looks to address the composition gap: an unauthorised outcome assembled entirely from authorised steps.

Presentation | Demo

DOWSERS

FRANCE
Dowsers

An AI Supervisory Agent powered by formal verification through the Dowsers Safety Oracle that provides safety scores for on-chain finance for ERC-20 tokens.

Presentation | Demo
KNF SupTech Team

POLAND
Polish Financial Supervision Authority (KNF)

A multi-agent continuous oversight system to identify and escalate potential MiCA Article 70 (safe keeping of client assets & funds) breaches.

Presentation | Demo
TRM Market Integrity Agent

US, UK
TRM Labs

An agentic surveillance system for supervision of predictive markets that scores contracts for market manipulation risk, monitors live markets, and escalates evidence-backed cases.
Presentation | Demo

Blue circular icon with a shield and exclamation mark, text reads 'AI- Driven Fraud & Scams'.

Combat the industrial-scale use of AI for criminal purposes to create deepfakes, synthetic identities, and hyper-personalised frauds and scams in order to protect consumers, financial systems and society.

Rwanda Anti Scam Team

RWANDA, UK, CANADA, NAMIBIA
Proto, FNA, National Bank of Rwanda, Bank of Namibia, GLEIF, Manchester University

Addresses the instant payments gap by using agents to report and recover funds of fraud and scam victims.

Presentation | Demo
FCAgents

UK
Financial Conduct Authority (UK)

A financial crime case augmentation & triage agent to support identification and prioritization of network-level investigations

Presentation | Demo
Project Zeus

DOMINICAN REPUBLIC
Superintendency of the Securities Market of the Dominican Republic (SIMV)

A proactive AI-powered honeybot for investment fraud detection that gathers evidence for enforcement.

Presentation | Demo

Agent Kiwi

NEW ZELAND
Financial Markets Authority of New Zealand

Turning reports into decision-ready intelligence, so regulators can disrupt scams before consumers are harmed

Presentation | Demo
Aviel

UK, UKRAINE
Aviel Technologies

Project Scout uses AI Honeybots to provide scam ad intelligence for proactive supervision by regulators.

Presentation | Demo
Drift Sentinel

ZAMBIA, NAMIBIA, SOUTH AFRICA
FSCA, UNISA, Guidehouse, Cenfri, CyprusCodes Akdemy Technologie

Closes the gap between emerging fraud and rulebook updates through a horizon-scanning agentic system that continuously stress-tests the crypto rulebook.

Presentation | Demo
The AI Whisper Room

MALAYSIA
Central Bank of Malaysia

Applies a whole of systems approach to tackling APP fraud through an Agentic Fraud Intelligence System that uses network graph analysis.

Presentation | Demo

Icon with a credit card and text 'Agentic Payments & Oversight'

Create oversight for high-velocity, machine-to-machine agentic payments and commerce, to detect and prevent automated illicit transactions, develop ‘Know Your Agent’ protocols, and agentic solutions for solving accountability and redress issues.

SUPPORTED BY ANT INTERNATIONAL

Team Axiom

PAKISTAN
State Bank of Pakistan

Supervisory agentic payments oversight that assigns persistent agent identifiers to identify systemic risks such as agentic herding.

Presentation | Demo
National Bank of Georgia

GEORGIA
National Bank of Georgia

Provides continuous oversight of AI payment agents, checking agent activity against what it was authorised to do.

Presentation | Demo

CEDE - AI Risk Transfer

SINGAPORE, UK
University of Edinburgh, Alan Turing, NTU, Imperial College

An oversight layer for agentic payments used to test whether a specific action remains within a user’s mandate to detect manipulation or execution mismatch.

Presentation | Demo
Golden Clock

SOUTH KOREA
Korea Financial Telecommunications and Clearings

Linking claims across institutions to the same mandate, agent action and execution, enabling a public authority to correlate evidence in disputed agentic transactions

Presentation | Demo
Ozone

UK, UAE
OzoneAPI

Overseeing agentic AI operating on open banking rails that enables visibility over "Know Your Agent" liability and risks of agent behaviour causing widescale errors and money laundering.

Presentation | Demo
National Bank of Ukraine

UKRAINE
National Bank of Ukraine

Examines how permissions, thresholds, authentication, confirmation and escalation mechanisms operate across different payment scenarios.

Presentation | Demo

Digital verification and DPI on a blue circular background with an ID card icon.

Develop agentic solutions, protocols or infrastructure to authenticate and verify AI agents and detect autonomous exploitation of digital public infrastructure layers (e.g. digital ID & data sharing).

Agent Reliability Profile

USA
ML Commons, Financial Services Working Group

A standardized, testable reliability label for AI agents to answer the question “Can I trust this agent?”

Presentation | Demo
Raidiam - Project Anchor

UNITED KINGDOM
Raidiam Services Limited

Project ANCHOR lets a bank prove which agent type was accredited, which agent instance acted, and whose regulated authority it carried, before a payment reaches the rails.

Presentation | Demo
Legis Labs

UK, SLOVENIA
Legis Labs, University of Ljubljana, University of Cambridge

An Agent Passport that gives banks a pre-execution authority check for AI-agent payments

Presentation | Demo

AgentReg-Pay

BANGLADESH
Bangladesh Bank

Helps authorities assess the resilience of Digital Public Infrastructure, starting with a Central Payment Platform (CPP)

Presentation | Demo
AstraSync

AUSTRALIA
Astra Sync AI

AstraSync is the root of trust for AI agents: verified identity, registration, runtime enforcement and on-chain anchoring.

Presentation | Demo
Future Native

USA, SOUTH KOREA, HONG KONG
Future Native Inc

Second Source gives a securities supervisor independent evidence of what a firm's AI trading agent did.

Presentation | Demo

Icon of a wallet with the text 'AI-Enabled Financial Advice' beneath it.

Mitigate risks to consumers using autonomous LLMs & agentic advisors for high-stakes decisions within financial services and the wider digital economy, including model biases, hallucinations, lack of explainability and compliance breaches.

SUPPORTED BY MONEYBOX

Probe

UK
Multiply Ai Limited

Uses an "Adversary" agent posing as realistic customers and a "Judge" agent to run thousands of test conversations against AI advice systems, flagging regulatory breaches.

Presentation | Demo
Agentic Alliance

US, EU
Independent

Spots emerging patterns in consumer complaints, runs supervisor-approved tests against AI chatbots to reproduce the suspected harm, and builds a traceable evidence pack.

Presentation | Demo

Agentic Pioneers

US, QATAR

A swarm of regulatory agents that stress-tests AI advisers in a sandbox and gives it a risk profile and trust label.

Presentation | Demo
Alethica

SCOTLAND
Alethica

Turns published regulatory rules into testable controls and runs deterministic checks across recorded AI advice journeys, showing which step failed and why.

Presentation | Demo
Naamse

US
UCLA, UC Davis, Amazon

An agent that sits in front of a compliance-locked AI, catches wrongful refusals, repairs the wording, sends the rest to a human, and logs every decision.

Presentation | Demo
OFAP

HONG KONG, SINGAPORE, UK
INSEAD, HKU, NUS, Chicago Global

An open standard that turns AI-drafted advice into a sealed record anyone can recheck offline to certify, refuse or escalate it.

Presentation | Demo