Your certification. Our infrastructure.

Kachyng does not hand you a SaaS product and wish you luck on your audit. We provision the environment your auditor needs — the right isolation tier, the right controls, the right evidence trail. You go to your auditor. You earn the certification.

Enterprise deals stall at compliance. Your buyer wants SOC 2. Their CISO wants HIPAA. Their legal team wants GDPR. Normally, you spend 6–18 months and $200K+ building infrastructure that can pass those audits — before you write a line of product code.

Kachyng is the infrastructure layer that makes your product compliance-ready from the day you provision an environment. Select the frameworks your customers require. We deploy the architecture that satisfies them. Your team focuses on product. Your auditor finds what they need.

01
Choose your compliance target

Tell us which frameworks your enterprise buyers require — SOC 2, HIPAA, PCI DSS Level 1, FedRAMP, GDPR, or all of the above. We map the requirement to the infrastructure tier it demands.

02
We provision the compliant environment

Kachyng deploys the isolation level your auditor needs — dedicated namespace, dedicated cluster, or a fully dedicated AWS account. Controls are instantiated from day one. Evidence collection starts automatically.

03
You walk into your audit ready

Your environment has the right isolation, the right encryption boundaries, the right access controls, and a documented control record. Your auditor reviews the environment. Your company earns the certification.

Every compliance framework has a minimum infrastructure isolation requirement. We map those requirements to four provisioning tiers. Customers select the tier that matches their audit target — we handle the rest. The provisioning behind these tiers is Kachyng's Sovereign AI infrastructure control plane.

Tier 1 · $
Multi-Tenant
JWT + schema isolation within shared infrastructure
Provisioning: Instant
SOC 2
CCPA
AML / OFAC
HIPAA
FedRAMP
Tier 2 · $$
Dedicated Namespace
Own Kubernetes namespace, dedicated database, dedicated KMS key
Provisioning: ~1 hour
SOC 2
PCI DSS L1
ISO 27001
HIPAA
GLBA
CCPA
FedRAMP
Tier 3 · $$$
Dedicated Cluster
Own EKS cluster, RDS instance, VPC, and KMS key — fully isolated stack
Provisioning: ~2 hours
All Tier 2
GDPR
FedRAMP Moderate
FedRAMP High
Tier 4 · $$$$
Dedicated Account
Dedicated AWS account via Control Tower — full compute and data sovereignty
Provisioning: ~4 hours
All lower-tier frameworks
FedRAMP High path
ITAR-adjacent

The frameworks your enterprise buyers require determine the infrastructure tier they need. Compliance is not one-size-fits-all — it is a product decision.

Customer TypeWhy It's Non-Negotiable
SaaS companyFirst enterprise contract requires it
Healthcare paymentsRequired for any PHI-adjacent payment flow
Regional bank or credit unionBoard-level compliance mandate
Large retailerCard brand requirement above 6M transactions/yr
EU enterpriseEU data location plus lawful international-transfer mechanisms (SCCs, adequacy)
Federal agency / contractorATO required before any federal procurement
Defense / intelligenceDedicated AWS account, no shared infrastructure, full audit chain

AI agents making payments on behalf of humans is a compliance surface that existing frameworks have not fully addressed. Kachyng is the only platform that has built answers into the infrastructure — not bolted on after.

How is an AI agent's spending authority defined?

Every agent operates under a human-granted delegation scope — amount limits, merchant categories, and time windows set at authorization. Agents cannot expand their own authority.

Who audits what an agent paid and why?

Every agent transaction writes to a signed, append-only audit log. Scope, session context, and originating human grant are recorded before settlement. Exportable for compliance review.

Can an agent's payment access be revoked instantly?

Yes. Delegation scopes are revocable in real time. A revoked agent cannot initiate or settle payments regardless of pending transactions.

What happens when an agent pays for the wrong thing?

Scope boundaries define what an agent is authorized to buy at grant time. Out-of-scope transactions are blocked at the gateway — not flagged after the fact.

AI agents making financial decisions on behalf of humans create a compliance surface that did not exist two years ago. Which requirements apply depends on the jurisdiction, the role Kachyng plays as provider or deployer, the decision the AI makes, the data processed, and the regulated status of the customer. Kachyng has built the controls into the infrastructure — not documented them in a wiki.

ControlImplementation
On-premises LLM executionAll LLM inference runs on Kachyng-managed infrastructure. No prompts, context, or customer data are sent to third-party AI providers. Data never leaves your compliance boundary.
Tamper-evident audit trail (immudb)Every LLM interaction — prompt, response, action taken — is written to an append-only, cryptographically verified audit log. Records cannot be altered or deleted after the fact.
Tenant-scoped LLM sessionsEach tenant operates in an isolated session scope. Prompts, conversation history, and inference context from one tenant are structurally inaccessible to another.
AI disclosure and acknowledgmentUsers are informed they are interacting with AI before the conversation begins. Disclosure must be acknowledged before the system accepts input. Acknowledgment is logged.
Human oversight gate for consequential decisionsWhen an AI agent proposes a payment, order, or contract action, execution is paused until a human reviews and explicitly authorizes it. The agent proposes. The human decides.
PII masking before LLM inputPersonal data is identified and masked before it reaches the language model. Three masking levels are available: full redaction, partial masking, and cryptographic hashing.

When AI agents initiate payments on behalf of humans, seven regulatory domains apply. No comprehensive regulation exists yet — Kachyng has built the compliance architecture ahead of mandates, mapping existing frameworks to agentic payment flows.

AreaHow Kachyng Handles It
Cardholder data protectionAI agents never touch raw card numbers. All payment credentials are tokenized via network token services (Visa VTS, Mastercard MDES). The agent runtime is descoped from PCI DSS Requirement 3. Agents operate as system accounts under Requirements 7.2.5.1 and 8.6 — unique credentials, least privilege, full audit logging.
Transaction authorizationAgent-initiated payments are classified as Merchant-Initiated Transactions under the Credential-on-File framework. The cardholder provides explicit, granular consent at enrollment — per-merchant, per-amount-range, per-category. Transactions carry COF indicators per Visa and Mastercard rules. High-value or first-time-merchant transactions trigger decoupled 3D Secure authentication on the cardholder’s device.
Automated contract formationThe Uniform Electronic Transactions Act Section 14 explicitly validates contracts formed by the interaction of electronic agents, even if no individual reviewed the agent’s actions at the time of the transaction. Kachyng’s consent architecture is built on this foundation: the user defines authority boundaries, the agent operates within them, and every action is logged.
Authority boundaries and liabilityAgent payments are structured as preauthorized transfers under Reg E § 1005.10(d), and users can stop individual payments. How Reg E’s unauthorized-transfer rules apply to an autonomous agent that exceeds granted authority is an evolving legal question — so Kachyng does not wait for it to settle: if an agent exceeds its delegated authority, Kachyng makes the consumer whole under its own contractual guarantee, enforced in code at the platform level.
Money transmissionKachyng operates as a payment facilitator — not a money transmitter. AI agents instruct licensed payment processors to execute transactions directly between cardholder and merchant. Kachyng never takes possession, custody, or control of user funds at any point in the transaction lifecycle.
EU operational resilience and accessibilityEU banks and payment institutions have applied DORA since January 2025 and push its ICT-risk, incident, audit, and exit requirements into vendor contracts — Kachyng supports these contractual controls. The European Accessibility Act (in force since June 2025) covers payment and identification flows: Kachyng checkout, authentication, and portal interfaces target WCAG-aligned accessibility.
US state privacy patchworkComprehensive state privacy laws now extend well beyond California — including Colorado, Connecticut, Delaware, Indiana, Kentucky, Maryland, Minnesota, Montana, Nebraska, New Hampshire, New Jersey, Oregon, Rhode Island, Tennessee, Texas, Utah, and Virginia. Kachyng provides configurable controls for consumer requests, sensitive-data consent, profiling records and opt-outs, decision explanations, and human-review workflows.
EU payment initiation (SCA)Strong Customer Authentication cannot be delegated to an AI agent. Kachyng uses decoupled SCA — the agent initiates the payment, the cardholder approves on their device via biometric or PIN. After initial authentication, recurring transactions with the same merchant use MIT exemptions per PSD2 rules.
Fraud classification and disputesAgent-within-authority disputes follow standard chargeback rules (cardholder vs. merchant). Agent-exceeds-authority disputes are covered by Kachyng’s contractual guarantee — Kachyng makes the consumer whole. Full decision audit logs (what the agent considered, what it chose, why) serve as compelling evidence under Visa Claims Resolution 3.0.

The legal foundation already exists.

UETA § 14 (Uniform Electronic Transactions Act, Section 14) states: “A contract may be formed by the interaction of electronic agents of the parties, even if no individual was aware of or reviewed the electronic agents' actions or the resulting terms and agreements.” This statute — adopted in 47 states — directly validates AI-initiated commerce. Kachyng's consent architecture, authority boundaries, and audit trail are built on this foundation.

Note to your General Counsel

If you are evaluating Kachyng for an enterprise deployment, your legal team will ask these questions. We have answered them here so your buyer can forward this page directly. Every answer reflects controls that are implemented and testable — not roadmap items.

Does customer data leave Kachyng infrastructure to reach an AI model?

No. LLM inference runs on Kachyng-managed compute. No data is transmitted to OpenAI, Anthropic, Google, or any third-party AI provider. This applies to all tiers.

Is there a tamper-evident record of AI decisions that affect financial transactions?

Yes. Every LLM interaction that leads to a consequential action (payment, order, contract) is recorded in immudb, an append-only database with cryptographic verification. Records are immutable. Each entry includes: timestamp, tenant, user, prompt hash, response hash, action taken, and outcome.

Can an AI agent autonomously execute a payment without human authorization?

No. The platform enforces a human oversight gate on all consequential decisions. When an AI agent proposes a financial action, execution is paused until a human reviews and explicitly authorizes it. This is a structural control, not a policy setting.

How is tenant data isolated in multi-tenant LLM deployments?

LLM sessions are scoped to the tenant at the infrastructure level. Session keys include tenant identifiers. There is no shared context window, session cache, or conversation state between tenants. This is tested and verified in CI.

Are users informed when they are interacting with AI?

Yes. The Colorado AI Act (SB 24-205, effective June 30, 2026) requires disclosure among broader duties — risk management, impact assessments, consumer notices, and human-review appeals for covered high-risk systems. Kachyng surfaces a disclosure notice at session start. The user must acknowledge it before the system accepts input. Acknowledgment is persisted and auditable.

What is the legal basis for processing personal data through the LLM?

PII is identified and masked before reaching the language model. Three masking levels are available: full redaction ([REDACTED]), partial masking (first/last character preserved), and SHA-256 hashing. The LLM operates on masked data. Original PII is not stored in AI logs.

Which AI-specific regulations does the platform address?
EU AI ActStaged: prohibited practices Feb 2025; GPAI Aug 2025; transparency Aug 2026; certain high-risk uses Dec 2027; high-risk in regulated products Aug 2028
Colorado SB 24-205Effective June 30, 2026 — risk management, impact assessments, consumer notices, discrimination controls, human-review appeals for high-risk AI
Texas TRAIGA (HB 149)Effective Jan 1, 2026 — prohibited AI uses, biometric restrictions, AI sandbox
California CCPA ADMT RegsRisk assessments from Jan 2026; automated decision-making tech rules phase in from Jan 2027
Connecticut CTDPA (PA 25-113)Amended July 1, 2026 — profiling transparency, automated-decision rights, inference access
UK Data (Use and Access) Act 2025In force June 2026 — modifies UK GDPR/PECR incl. automated decision-making rules
Korea AI Basic ActEffective Jan 22, 2026 — transparency and labeling for generative/high-impact AI
CFPBGuidance on AI chatbots in financial services
GDPR Art. 5Data minimization, purpose limitation, integrity
GDPR Art. 13Transparency and information obligations
GDPR Art. 22Automated decision-making and profiling
SR 11-7US financial regulatory guidance on model risk management
Does the AI agent ever touch raw cardholder data (PANs, CVVs)?

No. All payment credentials are tokenized via network token services (Visa VTS, Mastercard MDES) before the agent sees them. The agent operates exclusively with tokens. The agent runtime is descoped from PCI DSS Requirement 3. A targeted risk analysis under PCI DSS 4.0 Requirement 12.3.1 documents this architecture.

Who is liable when an AI agent exceeds its delegated payment authority?

Kachyng. If an agent initiates a payment outside the user’s defined authority boundaries (amount, merchant category, frequency), Kachyng makes the consumer whole under its contractual guarantee. How Reg E’s unauthorized-transfer framework (§ 1005.2(m), § 1005.6) applies when a delegated software agent exceeds granted authority is still an evolving legal question — Kachyng’s guarantee does not depend on its resolution, and the guarantee is enforced in code: the agent’s API call is rejected at the platform level if it exceeds scope.

Is this money transmission? Does Kachyng need an MSB license?

Kachyng is structured to avoid money transmission. It operates as a payment facilitator: AI agents instruct licensed payment processors to execute transactions directly between cardholder and merchant, and Kachyng never takes possession, custody, or control of user funds. FinCEN determines money-transmitter status on the facts and circumstances, and its 2014 administrative rulings on payment processors that do not hold funds set out the conditions Kachyng’s architecture is designed to satisfy.

What legal basis supports automated payment delegation without per-transaction human review?

UETA § 14 (Uniform Electronic Transactions Act, Section 14): “A contract may be formed by the interaction of electronic agents of the parties, even if no individual was aware of or reviewed the electronic agents’ actions or the resulting terms and agreements.” This is the statutory foundation. The user’s configuration of authority boundaries constitutes a manifestation of actual authority under Restatement (Third) of Agency § 2.01.

How does Strong Customer Authentication (SCA) work when an AI agent initiates an EU payment?

SCA cannot be delegated to an AI agent. Kachyng uses decoupled SCA under PSD2 Article 97: the agent initiates the payment, then the cardholder receives a push notification and authenticates on their own device (biometric or PIN). The agent does not perform authentication. After initial SCA, recurring transactions with the same merchant use Merchant-Initiated Transaction exemptions.

We are building toward our own certifications.

Kachyng is currently in the process of pursuing SOC 2 Type II, PCI DSS Level 1, HIPAA, and ISO 27001 certifications for its own platform. The infrastructure, controls, and evidence framework are in place. Certification timelines follow audit scheduling.

In the interim, we share full architecture documentation, control matrices, and third-party security assessments with enterprise customers under NDA. If your procurement team has a compliance questionnaire, send it to us — we will complete it directly.

To request a private security review, access architecture documentation, or discuss compliance requirements for a specific deployment: security@kachyng.com