News

Agentic Identity: An Interview with Cham Ho, Product Lead at AIR

AIR

September 9, 2026

TL;DR: In a HackerNoon interview on agentic identity, Cham Ho, product lead at AIR, set out why personal data a platform stores is turning into a liability, while a specific, consented and revocable claim from a trusted issuer, held as a verifiable credential, can become an asset. With eIDAS 2.0 requiring every EU member state to offer a digital identity wallet by December 2026, enterprises have stopped asking what decentralised identity is. They now ask how to prove age without collecting a document, and how to authorise an AI agent without giving it unlimited access. On agentic identity, Cham separates two problems: agent identity is still open research, while agent authorisation can ship today as operator-linked delegation.

Read the full interview: Your Passport Scan Is a Liability: Moca Network on Identity for Humans and AI Agents, by Ishan Pandey, HackerNoon, 9 September 2026.

Why identity reached the boardroom

Two pressures arrived together. Regulation has made holding personal data more expensive, with GDPR, India's DPDP Act, Indonesia's PDP law and the UK Online Safety Act all pushing companies to justify what they keep. Meanwhile HackerNoon cites deepfake volumes rising from about 500,000 to a projected 8 million in two years, with human detection now close to chance.

Cham's read is that identity has moved beyond compliance plumbing into product, risk and, eventually, revenue. Large technology platforms are building wallets, passkeys and device-bound credentials, but mostly inside their own ecosystems. The choice for an enterprise is whether identity becomes one more platform dependency or a portable trust layer that works across apps, partners and agents.

eIDAS 2.0: plan for the direction, not the date

Fewer than a third of member states are assessed as ready for the December 2026 deadline set by eIDAS 2.0. Cham does not think strategy should hinge on whether that date slips. The shift that matters is in what people expect: once someone can prove a fact through a national wallet, a bank or a telco, a new platform asking for yet another passport photo starts to look out of date.

Being ready, as Cham frames it, means a business can answer five questions:

Which claims do we actually need?
Which issuers do we trust to make them?
How will users present them?
How do we verify and revoke them?
How much raw personal data can we stop storing?

For how those claims are defined under EU rules, see our explainer on qualified electronic attestations of attributes.

Where the onboarding money goes

HackerNoon puts conventional onboarding at roughly €70 to €100 per user once document capture, liveness, sanctions screening, manual review, storage and drop-off from failed uploads are counted, against €3 to €8 with wallet-based credentials. Cham cautions against treating any single KYC figure as universal, since markets and providers differ. The direction holds regardless: the conventional model pays for trust every time it is needed.

A credential does not eliminate the first trusted check. What it eliminates is the repetition. Once a claim such as KYC passed or over-18 exists, the next platform checks a proof instead of collecting the document again. The part teams tend to underestimate is operational: choosing claims that are useful, selecting issuers, and handling freshness, expiry, revocation and user experience.

How issuance and presentation work

Party	Role	What it holds
Issuer	A bank, telco, KYC provider, loyalty programme or marketplace decides which claim it will stand behind and signs it	The source data
User	Receives the credential encrypted to them and approves each presentation with their private key	The credential
Verifier	Requests one specific proof, such as over 18 or KYC passed	Only that proof, not the birth date or document behind it

Personal data stays off-chain. The chain and verification layer handle trust, status, proof verification, revocation and settlement. That is a structural change to breach exposure rather than an incremental one: under the old model, every verifier becomes another archive of passports and selfies.

Held data as a liability, a claim as an asset

The commercial model follows from those mechanics. The issuer earns from trust it has already established, the verifier pays for only the proof it needs, and the user skips another onboarding flow. Cham is equally clear on the limits. A platform cannot relabel its database as a credential network, and the model fails without verifier demand, a trusted issuer, and a reason for users to present the claim. In Cham's words, "a raw database row has limited value."

Verification at the source, not detection alone

Deepfake detection is an arms race, and Cham does not present a single proof-of-humanity check as the answer. The stronger signal comes from parties attesting to what they genuinely know: a bank to KYC status, a telco to account tenure, a marketplace to seller reputation. Each is imperfect on its own, but together they are far harder to fake than one selfie. Cham is explicit that this complements fraud systems and does not address a brand-new unknown user, a compromised issuer or social engineering.

Agentic identity: why agent identity and agent authorisation are different problems

Problem	Status	Why
Agent identity	Open research	People have durable anchors such as biometrics, government ID, financial history and long-held accounts. An agent's model, tools or capability can change in a click, so it is unsettled whether its identity is the model, wallet, developer, operator, policy or runtime.
Agent authorisation	Shippable today	The operator, a person or organisation, already holds identity and permissions. The task is extending that authority to an agent for a defined context, scoped, consented and revocable.

Several questions remain open in Cham's account: agent-to-agent negotiation, shared standards, proving an agent stayed within its bounds across many steps, and where liability sits when a user, an agent, a model provider and a platform all touch one transaction. An agent holding a key does not, on its own, give it an identity.

For the verification side of the same problem, see our Know Your Agent (KYA) guide and our analysis of agentic payments.

What production at SK Planet taught

According to HackerNoon, AIR Identity is integrated across roughly 29 million users and 95,000 merchants at SK Planet, operator of OK Cashbag, South Korea's largest integrated rewards programme. Cham's lessons from running at that scale are practical. A specification is only a starting point, and real Android devices surface journeys that break in ways no spec predicts. Most users do not think about wallets or sovereignty, so chain, gas and seed phrases stay hidden and the experience reads as a feature of an app they already use.

Consent proved the harder design problem. Few people expand permission details when the incentive is attractive, a pattern already visible with agents connected to email and calendars. The balance Cham describes is identity design that stays simple, scoped and revocable, and almost invisible until the user genuinely has a decision to make.

What to do this quarter

List the personal data you hold and ask whether the business really needs to keep it.
For each field, ask whether you need the raw value or only a proof derived from it, such as over 18 instead of a full date of birth.
Pick one flow and replace the raw data with the narrowest proof possible, backed by a trusted issuer.
Work with providers that handle standards such as OID4VP rather than rebuilding wallet and credential infrastructure in-house.

Where AIR fits

AIR Identity provides the rails described in the interview: the AIR Account, encrypted credential infrastructure, verification programs and SDKs. The issuer remains the source of trust. A credential issued to a user's own AIR Account is encrypted to that user and moves only when they approve it, and an enterprise that has already completed KYC, onboarding or a membership check can issue to that user without a separate wallet step. Verifiers receive the proof they asked for, with minimal data custody, and growth teams can acquire users who already hold the credentials their product requires.

AIR extends the same credentials to AI agents, so an agent can prove who it acts for and what it is authorised to do.

Frequently asked questions

What is agentic identity?

Agentic identity covers how an AI agent proves who it acts for and what it is allowed to do. In a HackerNoon interview, Cham Ho, product lead at AIR, splits it into two parts: agent identity, which remains open research, and agent authorisation, which can ship today as scoped, consented and revocable delegation from a person or organisation.

What does eIDAS 2.0 require by December 2026?

Every EU member state must offer citizens a digital identity wallet by December 2026. Cham's advice is not to wait for every state to be ready: map the personal data you collect, then pilot one widely useful claim, such as over-18, KYC passed, residency or loyalty tier, as a credential this quarter.

How does selective disclosure reduce breach risk?

The issuer keeps the source data and signs a specific claim. A verifier that only needs to know a user is over 18 receives that proof, not the birth date or the passport scan, so it does not become one more store of identity documents that can be breached.

How can an AI agent act for someone without unlimited access?

Through operator-linked delegation. The person or organisation behind the agent already holds identity and permissions, and extends a bounded slice of that authority to the agent for one context, with consent and the ability to revoke it. A credential lets the agent prove who it acts for without being given access to everything.

Where is AIR Identity running in production?

According to HackerNoon, AIR Identity is integrated across roughly 29 million users and 95,000 merchants at SK Planet, the operator of OK Cashbag, South Korea's largest integrated rewards programme.