Arbiter

Decision, safety, and infrastructure for the work that has to be right.

A multi-industry company building decision, safety, and compliance infrastructure across five regulated and high-stakes domains.

5
regulated & high-stakes domains
Active
development across every domain
2026
founded, Republic of Korea
Open
select repositories public
FIG. 01Finance

A decision a regulator can replay.

Building neutral trust and compliance infrastructure for money, autonomous agents, and cross-border trade — wherever a high-stakes call must be made and later proven. Each decision produces a signed, tamper-evident record an examiner can verify independently: who decided what, on whose authority, and that the record has not changed since.

ForStablecoin issuers, banks, operators of autonomous AI agents, cross-border traders, and the regulators who audit them.
ProductCaravanA neutral trust rail for cross-border trade.ProductTollgateSanctions screening decisions an examiner can replay offline.
FIG. 02Agriculture & Food Security

Cold storage that reaches the last village.

Developing solar-powered cold-storage infrastructure for the places the electrical grid does not reach. Up to a third of a smallholder harvest spoils before it can be sold, for want of refrigeration. The work is a pay-as-you-go cold chain at village scale, so a farmer’s crop becomes income instead of waste.

ForSmallholder farmers in sub-Saharan Africa, and the food-security programs that serve them.
ProductHarvestGuardA solar-powered, pay-as-you-go cold chain for smallholder farmers.ProductFarmPilotAn AI copilot that turns a smartphone into a pocket agronomist for smallholders.
FIG. 03Cybersecurity & Fraud

Tell a real voice from a synthetic one.

Developing systems that detect cloned and synthetic voices in real time, to stop the fastest-growing fraud of the decade: the scam phone call. The work spans the whole kill-chain — verify the voice is human, defend the call while it is happening, and help victims recover afterward — and runs on the device, so the audio never has to leave it.

ForEvery phone user a scam can reach — starting with the elderly, who are targeted most and recover least — and the banks and fintechs that carry the loss.
ProductHALOOn-device defense that shields elders from voice-phishing, in real time.ProductGoldentimeA recovery copilot for the first hour after a voice-phishing transfer.
FIG. 04Health & Bio

Catch the error before the claim is denied.

Two efforts. The first reads the documents healthcare runs on before a human has to, catching the errors that get a claim denied at submission rather than months later. The second is early-stage research toward giving a voice back to people who have lost the ability to speak, through a silent-speech neural interface.

ForU.S. healthcare providers and the billing companies that serve them; and, in research, people living with speech loss.
FIG. 05AI Research & Safety

Where capability and calibration diverge.

Research into a counter-intuitive failure of large models: a system can grow more capable at a task while growing worse at knowing when it is wrong. The work is to measure that gap rigorously, and to harden models against the structured perturbations that exploit it.

ForThe AI-safety and model-evaluation field, and anyone deploying models where a confident wrong answer is dangerous.

The company

Five domains. One standard of proof.

Arbiter builds decision, safety, and infrastructure for regulated and high-stakes domains: places where being wrong is expensive and someone is required to be right.

Every product is held to the same standard. It has to work under scrutiny, hold up when the record is examined years later, and state plainly what it can and cannot prove.

The company was registered in 2026, after the work was already underway. Each product was researched, engineered, and tested as working software before there was an entity to hold it. Nothing here began as a business plan waiting for a build.

The unifying idea is the same in every domain. Take an ambiguous input, apply the rules, make the call, and keep proof a third party can check.

Leadership

Dayeon Kang

I’m the founder of Arbiter, a developer, quantitative researcher, and civic-technology builder. I lead design, engineering, and delivery across all five domains, from the cold-chain hardware thesis to the cryptographic trust rails to the AI-safety research.

My method is the same everywhere: find a place where a high-stakes decision is made badly or slowly, build the system that makes it well, and keep proof it was right.

dayeon603@gmail.com