Security

Why this page exists

A broker embedding Zeig is asking a narrower question than "is the site encrypted."

Clients will type positions, account context, and intent into a conversational interface that looks like yours. That interface routes through an orchestrator and specialist agents. Those agents call frontier models. The security problem is not only transport encryption. It is whether a query from one book can leak into another, whether a model provider can train on your clients, and whether an agent can act outside the rules you set.

This page is the answer to that question.

Zeig is infrastructure. Your name stays on the glass. Your relationship stays yours.

How we handle AI security

Zeig is not a chatbot bolted onto a terminal. An orchestrator routes each question to specialist agents, then grounds the answer in live market data. That architecture is also how we contain risk.

Bounded agents

Agents analyse, retrieve, and recommend. They do not hold client funds. They do not place live orders today. When execution connectivity ships, it will sit behind broker-controlled permissions, not inside an unconstrained model loop.

You configure behaviour at the orchestrator: risk posture, which instruments can be discussed, whether trade setups are shown, and the compliance language that travels with every answer.

No training on your book

Queries, conversation history, account context, and positions are not used to train Zeig models. They are not used to train the third-party models we call. Analysis is produced for the session that asked for it, not absorbed into a shared brain.

Least context, not maximum context

The orchestrator sends an agent the context required to answer the question: instrument, timeframe, and the relevant account fields you have chosen to share. It does not send an open dump of the client file. Position awareness is a feature you switch on. It is not a default vacuum.

Isolated inference

Each user's queries are processed in isolation. One client session cannot read another. One broker tenant cannot see another broker's users, prompts, or configuration.

Encrypted calls to model providers

Zeig uses multiple frontier models. Every call to a provider is authenticated and encrypted. Providers receive what they need to generate a response. They do not receive a licence to retain, reuse, or train on that traffic under our contracts.

Prompt and output controls

User text is treated as untrusted input. Orchestrator instructions, broker policy, and system prompts are separated from the client's question so a user cannot talk an agent into dropping policy, revealing another tenant, or changing its role. Outputs that recommend action carry the disclaimer and constraints you configure.

Human-readable audit trail

Broker deployments can inspect what was asked, which agent handled it, which data sources were used, and what was returned. That log is for your compliance team, not a marketing dashboard.

What never happens to a broker's users

• We do not sell personal data, queries, or analysis.
• We do not train on client conversations.
• We do not mix tenants.
• We do not ask end clients for brokerage passwords when Zeig runs inside your client area. Authentication stays with you.
• We do not custody money. Zeig is an intelligence layer, not a wallet or a prime.
• We do not silently change orchestrator rules. Your configuration is yours.

Isolation

Security for a white-label book is tenancy, not just TLS.

Broker from broker. Each deployment is a separate tenant: branding, orchestrator rules, allowed instruments, data connections, and user directory.

Client from client. Sessions and query histories are scoped to the authenticated user. Account context attached to a request cannot be retrieved by a different login.

Zeig from your stack. In a broker integration, Zeig consumes the identity and market context you pass over authenticated APIs. It does not become the system of record for accounts, balances, or legal onboarding.

Analysis from execution. Research agents can prepare a setup. They cannot move cash. Execution, when it exists, is a separate, permissioned path.

Direct platform

Accounts use OAuth 2.0, hashed passwords, automatic session expiry, and secure token handling. Two-factor authentication is available.

Inside a brokerage

The intended pattern is broker authentication. The client signs in to you. You assert identity and the account context Zeig is allowed to see. Zeig does not need a second password from the trader, and it does not store the credentials they use to reach your platform.

Inside Zeig

Staff access to production systems is role-based, logged, and limited to what the role requires. There is no standing open access to client query stores.

Encryption and infrastructure

In transit. TLS 1.3 on connections between the client, Zeig, and upstream providers.

At rest. AES-256 on stored data, the same class of cipher used by governments and financial institutions.

Hosting. Enterprise cloud providers with SOC 2 Type II reports. Network controls include layered firewalls and DDoS mitigation.

Operations. 24/7 monitoring, automated threat detection, periodic assessments, and penetration testing.

Encryption is the floor. It is not the argument for putting Zeig in front of clients. Isolation and AI policy are.

Privacy, retention, and GDPR

We collect what the product needs to run: identity needed to authenticate, queries needed to answer, and the account or market context you choose to attach. Nothing else is harvested "in case it is useful later."

• Users and brokers can request deletion. We honour complete erasure of stored personal data and query history subject to any short, documented legal hold.
• Data portability is supported where GDPR requires it.
• Consent and processing notices can sit in your client area. In a white-label deployment the broker is typically the controller for end-client personal data; Zeig acts as processor under a DPA.
• Practices are designed against UK GDPR and EU GDPR. Cyprus and UK entities are the contracting companies.

If you need a DPA, subprocessor list, or a review of what leaves your region, write to info@zeig.ai.

Responsible disclosure

If you find a vulnerability, send it to info@zeig.ai. Do not use it against production clients.

We will:

• Acknowledge the report within 48 hours
• Keep you updated while we investigate
• Not pursue legal action against good-faith researchers
• Credit you if you want the credit

Questions compliance and security teams ask

Does Zeig train on our clients?

No. Not internally, and not via the model providers we call.

Who sees a client's positions?

Only the session that is allowed to see them, and only if you have enabled account context for that integration. Other tenants cannot.

Can an agent place a trade on its own?

Not today. Agents analyse and prepare. Live execution is a separate, forthcoming control surface, not an unsupervised model action.

Where do prompts go?

To Zeig's orchestration layer, then to the model chosen for that task, over authenticated TLS. Providers are bound not to train on that traffic.

Do you sell query logs?

No.

Can we keep our own login?

Yes. White-label is designed around broker authentication so the client never creates a parallel password at Zeig.

Are you a regulated broker?

No. Zeig Labs Ltd (UK) and ACM 101 Solutions Ltd (Cyprus) build the platform. You remain the firm your clients have the account with. Your regulatory perimeter does not transfer to us because the interface uses our agents.

Can we put our own risk language on answers?

Yes. Compliance disclaimers and orchestrator rules are part of the broker configuration, not a hidden Zeig default.

Talk to security

Questions about architecture, a DPA, or a deployment review: info@zeig.ai

Integrating Zeig into a client area: use the Talk to sales form, or write to the same address.