AMBR security and controls: data access, approvals, testing and incidents for Ambre and MIA
This AMBR security page sets out, in six dated sections, how Ambre and MIA are controlled: which model runs them, what data they can read or change, which actions need a person, how they are tested, what incidents have occurred and where to check status. Where AMBR has not yet published a fact, the section says so with the date it was last checked.
Not financial advice. Crypto is highly volatile and may drop in value significantly. You may lose the amounts you invest and your investments do not benefit from any form of financial protection.
Reviewed by Head of Compliance. Company site read on 6 October 2026; registers read and this page last checked on .
| Section | What AMBR has published | As of |
|---|---|---|
| 1. Model and provider | Not yet published | |
| 2. Data access | In outline: three account types for Ambre and six sources for MIA. Connection methods and write access are not yet published. | |
| 3. Permissions and approvals | Three statements on who leads, who decides and who executes. Limits per action are not yet published. | |
| 4. Testing and evaluation | Not yet published. Two operating figures for MIA, with no method. | |
| 5. Incident log | No incident has been published. | |
| 6. Status and availability | No status page exists. |
People who ask whether an AI agent for personal finance is legitimate are usually asking the six questions above. This page answers them for AMBR's own agents only, with dated facts. General explanations of how agents are kept within bounds are in the guide to risk controls for AI agents that touch digital assets, and the glossary defines guardrails, human in the loop and other terms.
1. Model and provider
As of
A trust centre for an AI agent starts with the model that does the work. AMBR has not yet published which model runs Ambre or MIA, who provides it, which version is in use, or whether what you type and connect is used to train models.
Until those four facts are published, treat each of them as unknown. They are fair questions to put to AMBR before you connect an account or a brand channel.
- Model
- Not yet published
- Provider
- Not yet published
- Version
- Not yet published
- Training on customer data
- Not yet published
2. Data access: read versus write
As of
Read access lets an agent see data. Write access lets it change something: place an order, move an asset, publish a post. The two carry different risks, so they are listed apart for each agent.
Ambre
Ambre reads the accounts you connect. The live Ambre page says Portfolio Intelligence works “across brokerages, exchanges, and wallets”. AMBR has not published which venues are supported, how a connection is made, or whether Ambre holds any write access to a connected account.
The tracker page keeps the connections table and explains how the AI portfolio tracker connects accounts. What happens to personal data is in the privacy policy.
MIA
MIA reads six monitored sources, as the live agents page lists them: a user's brand, competitors, markets, social channels, news and AI search. Nothing AMBR has published describes MIA connecting to a brokerage, an exchange or a wallet.
What MIA can change, such as a post, a message or a budget, depends on approvals. AMBR has not published which of those MIA can do without a person, so section 3 lists it as open.
| Agent | Data read | Data written | Stored where |
|---|---|---|---|
| Ambre | Portfolio data from the accounts you connect, across brokerages, exchanges and wallets. The venue list is not yet published. | Not yet published | Not yet published |
| MIA | Signals from six sources: a user's brand, competitors, markets, social channels, news and AI search. | Not yet published. AMBR describes the output as “approved actions”. | Not yet published |
Source: the live Ambre and agents pages, read 6 October 2026. Last reviewed .
3. Permissions and approvals
As of
AMBR has published three statements about who leads, who decides and who executes. They are quoted in full below. It has not published the limits that apply to each action.
approval model, as published by AMBR
- led by
- “AI leads the experience while you remain in control. Human expertise is available when judgment or execution is required.”
- decided by
- You. Ambre works “with users remaining in control of consequential decisions and actions”.
- executed by
- “Where you choose to pursue a transaction or other financial activity, relevant services may be provided within AMBR's licensed scope or through applicable third-party providers.”
- limit
- not yet published
Source: the live Ambre and agents pages, read 6 October 2026.
The table below is the same one that appears on the Ambre page, which also sets out what Ambre can and cannot do. For MIA, the product page explains how MIA prepares work for approval.
| Action | Who acts | What AMBR has published |
|---|---|---|
| Read balances and positions | Ambre alone, on the accounts you connect | Portfolio Intelligence works across brokerages, exchanges and wallets. The connection method is not published. |
| Raise a signal | Ambre alone | Personalized Market Signals is a named feature. A signal is information. |
| Create a monitoring task | Needs your approval | Ongoing Monitoring and Tasks is a named feature. Who sets up a task is not published. |
| Prepare a transaction | Needs your approval | AMBR says users remain in control of consequential decisions and actions. |
| Execute a transaction | Done by a licensed entity or a third party | AMBR says human expertise is available when judgment or execution is required, and that services are provided within its licensed scope or by third-party providers. |
| Change connected accounts | Needs your approval | Not published. |
Source: the live Ambre and agents pages, read 6 October 2026. Last reviewed .
AMBR's statements settle two rows: preparing a transaction and executing one. The other four rows describe the behaviour to expect and have not been confirmed by AMBR. If an approval fails or an agent acts on wrong information, the guide on liability when an AI agent acts on your money explains how responsibility is split.
4. Testing and evaluation
As of
AMBR has not yet published how it tests Ambre or MIA. A section like this one should carry four things: the test sets used, the red-team exercises run, the date of the latest evaluation and its result.
Two public references show what such testing covers. The OWASP Top 10 for LLM Applications 2025 lists prompt injection (LLM01:2025) and excessive agency (LLM06:2025) among its risks. The NIST AI Risk Management Framework (AI RMF 1.0), released on 26 January 2023, is intended for voluntary use. AMBR has not said that it follows either.
For MIA, AMBR has published two operating figures. They describe use inside its own marketing businesses, iClick and CMRS, and they are not test results. The baseline, the measurement period and the definition of a workflow are not published.
mia, operating figures as published by AMBR
- figure
- more than 1,250 automated workflows run
- figure
- turnaround times cut by an average of 85% within four months of rollout
- stated by
- AMBR; not independently verified
- source
- rebrand release, 31 August 2026
- as of
- checked 11 October 2026
5. Incident log
As of
An incident here means an event in which an agent, or the systems behind it, exposed data, acted outside its permissions or was unavailable in a way that affected users.
| Date | Agent | What happened | What changed afterwards |
|---|---|---|---|
| No incident has been published as of 11 October 2026. | |||
No industry standard exists for disclosing incidents that involve an AI agent, so the notice period will be AMBR's own commitment once it is set. It has not been set yet. When an entry is added, it will give the date, the agent, what happened, who was affected and what changed afterwards.
The risks that sit behind such incidents, model error and market risk among them, are set out in the risk disclosure for AI agents and virtual assets.
6. Status and availability
As of
No status page exists on the company's host, so there is no public record of uptime or outages for Ambre or MIA. When a status page is published, its address will be listed here and in the change log below.
- Status page
- Not yet published
- Uptime history
- Not yet published
- Ambre
- Available by invitation, with a public waitlist at h5.ambre.ai
- MIA
- Runs on its own product site at mia.ambr.io
What the agents cannot do
The limits are stated in the same words as on the product pages. For Ambre they are plain statements about the agent as it stands today. For MIA only one line is published, and the rest is marked as open.
Ambre
Six limits on Ambre
- Ambre does not give personal investment advice.
- Ambre does not promise any outcome, and nothing on this page is a forecast.
- Ambre does not move assets on its own decision: you decide, and a licensed entity or a third-party provider carries out any transaction.
- Ambre does not replace a licensed adviser.
- Ambre is not open to everyone: access is by invitation only.
- Ambre's outputs can be wrong or out of date, because it is an AI system.
MIA
What is published about MIA
AMBR describes what MIA produces as “marketing intelligence, structured work, and approved actions”. Whether MIA can publish content, send a message or spend a budget without a person's approval:
Not yet published
MIA is an AI system. Its outputs can be wrong. It prepares marketing work for approval; review each output before it is published or acted on. MIA is a marketing product and does not give investment advice.
Certifications, audits and insurance
The company site publishes no certification, audit report or insurance statement, so this page claims none. Each row below is the place where an issuer, a scope and a date will go once the owner supplies them.
-
Custody model
who holds client assets
AMBR has not published who holds assets for clients of its licensed entities, or whether Ambre ever takes custody of anything. The VARA register entry for Amber Premium FZE (VL/26/03/003) lists three activities, and Custody Services is not one of them. Amber Premium FZE is permitted to serve Institutional Investors and Qualified Investors.
Not yet published issuer: none named
VARA public register entry, read 11 October 2026 -
Key management
credentials for connected accounts
How credentials for connected accounts are created, stored and revoked is not published. That covers exchange API keys, brokerage logins and wallet addresses, and whether any key that Ambre holds could trade or withdraw.
Not yet published issuer: none named
as of 11 October 2026 -
Certifications and audits
issuer, scope, period
No security certification, audit report or penetration-test statement appears on any page of the company site. When one exists, this row will name the issuer, the scope, the period covered and the report date.
None published issuer: none named
company site read 6 October 2026 -
Insurance
carrier, limit, scope
No insurance statement is published: no carrier, no limit and no scope. This page claims no insurance cover, and the warning at the top of the page applies as written.
None published issuer: none named
company site read 6 October 2026 -
Incident history
dated record
The record is the incident log in section 5. It is empty, and the security owner has yet to confirm it.
No incident published issuer: AMBR
as of 11 October 2026
Report a security problem
AMBR has not yet published a security contact address or a disclosure policy. Until it does, use the contact page to report a problem or contact support.
Before you answer a message that claims to come from AMBR, check the sender against the official domains, which the contact page also lists. The About page names the company and its licensed entities.
- Security contact
- Not yet published
- Official domains
- www.ambr.io, mia.ambr.io, ir.ambr.io and h5.ambre.ai
- Media relations
- +65 6022 0228. This line is for the press, not for security reports.
Change log
Every change to a model, a permission or a data source gets a dated entry here within 7 days. The first entry is the publication of this page.
More background is in the AMBR guides to how AI agents work and how they are controlled.
- Page published with six dated sections. No change to a model, a permission or a data source has been recorded.
The controls to ask any provider about
The guide lists what to check before an agent gets near a wallet or an exchange account. It is for information only and is not investment advice.