# Trust and security: who secures what, layer by layer

> What you secure, what Fantasti secures and what the infrastructure layer secures, with a status on every control. Private Preview.

Source: https://fantasti.ai/trust
Reviewed: 2026-10-13
Platform: Private Preview. We are not taking access requests yet. Ask us a question, or propose a partnership.

Responsibility · Controls · Sub-processors · [Private Preview] · Reviewed 2026-10-13

## Who secures what, layer by layer.

The Fantasti Cloud Platform is a control plane. The infrastructure layer under it is operated by infrastructure providers. This page separates the two, gives every Fantasti control a status, and states what a provider must hold before a workload runs on it.

[Read the responsibility table](https://fantasti.ai/trust#responsibility) [Report a vulnerability](mailto:security@mail.fantasti.ai?subject=Vulnerability%20report)

State on 2026-10-13 Opens with the first cohort. No account is open yet. Every control on this page is described as it is being built for that cohort.

Responsibility stack. Source · Fantasti. Reviewed 2026-10-13.

*DWG 01 · Responsibility stack · Source · Fantasti · Reviewed 2026-10-13*

Three planes, top to bottom: you, Fantasti and the infrastructure layer. Each of the 17 rows of the shared responsibility table is drawn once, under its number: on a card inside a plane when that layer answers for it alone, on a plate across a joint when two layers share it, and in the well under the stack when every layer has a part. Each plane ends on its independent assurance: You, operated by You: Yours; Fantasti, operated by Fantasti: NOT YET AUDITED; Infrastructure layer, operated by Infrastructure providers: NOT YET NAMED Connections: You to has a part; Fantasti to has a part; Infrastructure layer to has a part.

A You Operated by You

B Fantasti Operated by Fantasti

C Infrastructure layer Operated by Infrastructure providers

Every layer has a part

- 01 Data you bring
  Classification, lawfulness, minimization
- 02 Your applications
  - Containers
  - Images
  - Model weights
- 05 Backups of your data
- Independent assurance
  Yours
  Your assessor evaluates your deployment.
- 08 Identity platform
  - Sign-in
  - SSO
  - MFA
  Through our identity sub-processor
- 09 Fantasti Cloud Platform (highlighted)
  - Console
  - API
  - MCP server
  - Orchestrator
  - Metering
  - Billing
  - operator: AI agents, by design
  - under: written rules and hard limits
  - entity: Fantasti Technologies LLC
- Independent assurance
  `NOT YET AUDITED`
  No audit period has started.
- 15 Hosts and network
  [Image: 8-GPU node, 8 × 400 Gb/s InfiniBand, line drawing]
  Hypervisor, host OS, network underlay, InfiniBand
- 16 Hardware
  Hardware lifecycle and media disposal
- 17 Data centers
  Physical and environmental security
- Independent assurance
  `NOT YET NAMED`
  Each is listed in §03 with what it holds, before a workload runs on it.
- 03 Users, roles and API keys
- 04 Network exposure
- 06 Monitoring and logs
- 10 Tenant separation
- 11 Encryption in transit
- 12 Encryption at rest
- 13 Vulnerability management
- 14 Incident response and notification
- 07 Compliance of your use

A You Operated by You

B Fantasti Operated by Fantasti

C Infrastructure layer Operated by Infrastructure providers

Every layer has a part

- 01 Data you bring
  Classification, lawfulness, minimization
- 02 Your applications
  - Containers
  - Images
  - Model weights
- 05 Backups of your data
- 07 Compliance of your use
- Independent assurance
  Yours
  Your assessor evaluates your deployment.
- 08 Identity platform
  - Sign-in
  - SSO
  - MFA
  Through our identity sub-processor
- 09 Fantasti Cloud Platform (highlighted)
  - Console
  - API
  - MCP server
  - Orchestrator
  - Metering
  - Billing
  - operator: AI agents, by design
  - under: written rules and hard limits
  - entity: Fantasti Technologies LLC
- Independent assurance
  `NOT YET AUDITED`
  No audit period has started.
- 15 Hosts and network
  [Image: 8-GPU node, 8 × 400 Gb/s InfiniBand, line drawing]
  Hypervisor, host OS, network underlay, InfiniBand
- 17 Data centers
  Physical and environmental security
- 16 Hardware lifecycle and media disposal
- Independent assurance
  `NOT YET NAMED`
  Each is listed in §03 with what it holds, before a workload runs on it.
- 03 Users, roles and API keys
- 04 Network exposure
- 06 Monitoring and logs
- 10 Tenant separation
- 11 Encryption in transit
- 12 Encryption at rest
- 13 Vulnerability management
- 14 Incident response and notification

- A You
  - 01: Data you bring
  - 02: Your applications, containers, images
  - and model weights
  - 05: Backups of your data
  - 07: Compliance of your use
  - operated by: You
  Independent assurance
  Yours
- Shared across the joint A and B
  - 03: Users, roles and API keys
  - 04: Network exposure
  - 06: Monitoring and logs
- B Fantasti Cloud Platform (highlighted)
  - 08: Identity platform
  - 09: Console, API, MCP server,
  - Orchestrator, metering, billing
  - operator: AI agents, by design
  - under: written rules and hard limits
  - entity: Fantasti Technologies LLC
  - operated by: Fantasti
  Independent assurance
  `NOT YET AUDITED`
- Shared across the joint B and C
  - 10: Tenant separation
- C Infrastructure layer
  [Image: 8-GPU node, 8 × 400 Gb/s InfiniBand, line drawing]
  - 15: Hypervisor, host OS, network
  - underlay, InfiniBand
  - 16: Hardware lifecycle and media disposal
  - 17: Physical and environmental security
  - operated by: Infrastructure providers
  Independent assurance
  `NOT YET NAMED`
  Each is listed in §03 with what it holds, before a workload runs on it.
- Every layer has a part
  - 11: Encryption in transit
  - 12: Encryption at rest
  - 13: Vulnerability management
  - 14: Incident response and notification

- The layer Fantasti operates
- Every layer has a part

A plate across a joint is shared by the two layers it laps. Numbers key to §02.

1. [§01 Layers](https://fantasti.ai/trust#layers)
2. [§02 Responsibility](https://fantasti.ai/trust#responsibility)
3. [§03 Infrastructure](https://fantasti.ai/trust#inherited)
4. [§04 Status](https://fantasti.ai/trust#status)
5. [§05 Controls](https://fantasti.ai/trust#controls)
6. [§06 Agents](https://fantasti.ai/trust#agents)
7. [§07 Defense](https://fantasti.ai/trust#defense)
8. [§08 Data](https://fantasti.ai/trust#data)
9. [§09 Incidents](https://fantasti.ai/trust#incidents)
10. [§10 Sub-processors](https://fantasti.ai/trust#sub-processors)
11. [§11 Scrutiny](https://fantasti.ai/trust#disclosure)
12. [§12 Q&A](https://fantasti.ai/trust#qa)

§01 · Layers · [Three layers]

### The three layers

- A You. Your code, containers, models and data. Who in your organization holds which role. What you expose to the internet. Whether your use meets the rules that apply to you.
- B Fantasti. The console, the API, the MCP server and the Fantasti Orchestrator. Identity and roles. The isolated cluster we build for you. Placement, metering and billing. The agents that operate the platform, and the written rules and hard limits set for them by Fantasti Technologies, the company behind Fantasti.
- C Infrastructure layer. Data centers, power and cooling, servers and GPUs, virtualization, the InfiniBand and Ethernet fabrics, block, file and object storage, and the encryption underneath them.

§02 · Responsibility · [Reviewed 2026-10-13]

### Shared responsibility, row by row.

responsible. shared. A blank cell means no role.

The infrastructure column is the part every agreement with a provider has to cover.

Row numbers are the numbers on [DWG 01](https://fantasti.ai/trust#top).

*TAB 02 · Shared responsibility · Source · Fantasti · Reviewed 2026-10-13*

**Shared responsibility: for each area, which of the three layers is responsible and which shares the duty**

| Row | Area | You | Fantasti | Infrastructure layer |
| --- | --- | --- | --- | --- |
| 1 | Data you bring: classification, lawfulness, minimization | Responsible | No role | No role |
| 2 | Your applications, containers, images and model weights | Responsible | No role | No role |
| 3 | Users, roles and API keys in your organization | Responsible. You assign and revoke | Shared. We provide sign-in, SSO, roles and key revocation | No role |
| 4 | Network exposure: ports, endpoints, policies you set | Responsible | Shared. We enforce the mode you choose | No role |
| 5 | Backups of your data | Responsible | No role | No role |
| 6 | Monitoring your workloads and reading your logs | Responsible | Shared. We supply metrics and logs | No role |
| 7 | Compliance of your use (sector rules, controller duties) | Responsible | No role | No role |
| 8 | Identity platform: sign-in, SSO, MFA | No role | Responsible. Through our identity sub-processor | No role |
| 9 | Console, API, MCP server, Orchestrator, metering, billing | No role | Responsible | No role |
| 10 | Tenant separation: your isolated cluster and its control plane | No role | Responsible | Responsible. Isolation of the hosts and the network underneath |
| 11 | Encryption in transit | Responsible. Inside your workloads | Responsible. TLS on Fantasti endpoints | Responsible. Its own service endpoints |
| 12 | Encryption at rest | Shared. Application-level, where your policy needs it | Responsible. We select encrypted storage configurations | Responsible. Storage encrypted at rest, as every agreement requires |
| 13 | Vulnerability management | Responsible. Your code and images | Responsible. Our code and managed images | Responsible. Platform, hosts, firmware |
| 14 | Incident response and notification | Responsible. Your workloads | Responsible. Single point of contact; we notify you | Responsible. Notifies Fantasti |
| 15 | Hypervisor, host OS, network underlay, InfiniBand | No role | No role | Responsible |
| 16 | Hardware lifecycle and media disposal | No role | No role | Responsible |
| 17 | Physical and environmental security | No role | No role | Responsible |

#### Where it depends on the product.

Four duties move with the product you run. Read across for yours.

*TAB 03 · Where it depends on the product · Source · Fantasti · Reviewed 2026-10-13*

**Where it depends on the product**

| Area | Instances | Clusters (isolated Kubernetes) | Workspaces, Sandboxes | Serverless, Managed MLflow |
| --- | --- | --- | --- | --- |
| Guest OS and node image patching | You | Fantasti | Fantasti (base images); you (your layers) | Infrastructure layer |
| Kubernetes control plane | n/a | Fantasti | Fantasti | Infrastructure layer |
| Container images and dependencies | You | You | You | You |
| Backups of your data | You | You | You | You |

#### Who does the hands-on work.

Infrastructure-layer incidents are worked with the operations team of the provider concerned.

[Planned]

*TAB 04 · Hands-on work on the infrastructure layer · Source · Fantasti · Reviewed 2026-10-13*

**Hands-on work on the infrastructure layer, and who does it**

| Work | Done by |
| --- | --- |
| Repairing and replacing hardware | The operations team of the provider that runs the hardware |
| Operating the data centers | The operations team of the provider that runs them |
| Reporting a fault and following it to repair | Fantasti: the Capacity team and the infrastructure desk, both AI agents |

Asking, reporting a fault and following up need no review. Anything that commits Fantasti to pay stays under the hard limits.

#### Who answers in law.

One legal person stands behind the Fantasti layer, whoever does the work.

Fantasti Technologies LLC is the legal person responsible for the service. It signs the contracts and the data-processing agreements, and answers legal process and regulators.

The service is designed to be operated by AI agents, under written rules and hard limits that Fantasti Technologies sets. [The limits they work under are in §06.](https://fantasti.ai/trust#agents)

These are the commitments Fantasti Technologies puts into the terms an account accepts before it opens. The terms are drafts, and where the two differ the terms govern.

1. A service that does what its documentation says.
2. One price of record: use is charged at the price list in force at the time of use.
3. Good faith in every decision about an account, with its reasons.
4. A security program on every product. It acts on records of use, never on the contents of your data.
5. A second look: a suspension, a closure or a withheld amount is reviewed on request, by a reviewer that did not make the decision.
6. Open to scrutiny: what has and has not been independently examined is stated, and security research done in good faith is welcome.

§03 · Infrastructure · [Reviewed 2026-10-13]

### The infrastructure layer, and what we require of it.

The infrastructure layer is operated by infrastructure providers under signed agreements. None is signed yet. Each provider is named here, with what it holds and a way to verify it, before a customer workload runs on its infrastructure.

*TAB 05 · Infrastructure providers, on record · Source · Fantasti · Reviewed 2026-10-13*

- Providers: Infrastructure providers NOT YET NAMED
- Agreements: To be signed
- Named here: Before a customer workload runs

**What Fantasti requires of every provider of the infrastructure layer, what is listed here once a provider is named, and how you verify it**

| Requirement | Fantasti requires | Listed here with each provider | How you verify it |
| --- | --- | --- | --- |
| 1 | ISO/IEC 27001 certification | The certificate and what it covers | On the provider's own page for the certificate |
| 2 | A SOC 2 Type II report | The report and how it is released | By request, from the provider that holds it |
| 3 | Encryption of storage at rest and in transit | Which storage is encrypted, and how | In the provider's published documentation |
| 4 | A data-processing agreement | The provider's legal entity, in the sub-processor table | In your Data Processing Addendum, once there is one |
| 5 | Data kept in the selected region | The regions on offer | On your quote, region by region |

Fantasti signs only with providers that meet all five. When a provider is listed, each row gives way to that provider's own certificate, report or contract term.

A provider's certificate covers the provider's systems. Fantasti's own status is in the next section.

#### Ask about a requirement.

Does your review depend on a standard that is not on this list? Tell us which one before an agreement is signed.

[Ask about a requirement](mailto:security@mail.fantasti.ai?subject=Infrastructure%20requirement)

§04 · Fantasti status · [Private Preview]

### Fantasti's own status, stated plainly.

Fantasti Technologies LLC holds no security certification and has not been through an independent audit. We are in Private Preview. When an audit starts, this table will name the standard, the auditor and the period.

*TAB 06 · Frameworks, Fantasti layer · Source · Fantasti · Reviewed 2026-10-13*

**Frameworks, Fantasti layer: each row with its status**

| Framework | Status | Note |
| --- | --- | --- |
| SOC 2 Type II | Not yet audited | No audit period has started. |
| ISO/IEC 27001 | Not yet audited | No certification audit has started. |
| Penetration test of the Fantasti layer | [Planned] | No third-party test has been completed. |
| HIPAA | Not supported | We do not sign business associate agreements during Private Preview. Do not send protected health information to Fantasti. |
| GDPR | [Planned] | Fantasti is designed to act as your processor for workload content. A Data Processing Addendum is not available yet. We publish the addendum before the first account opens. |
| CCPA | Stated | We do not sell personal data. |
| DORA · NIS2 | Not yet audited | Not assessed. Financial-sector customers who need contractual terms should contact us before signing. |
| Card data | [Planned] | Card details are entered with the payment processor and never reach Fantasti. |

§05 · Controls · [Reviewed 2026-10-13]

### Controls of the Fantasti layer, each with a status.

PRIVATE PREVIEW means it opens with the first cohort. EARLY ACCESS means it is being built for the first cohort. PLANNED means it is on the roadmap, with no date. INHERITED means the infrastructure layer provides it.

- [Private Preview]: 05
- [Early access]: 07
- Inherited: 01
- [Planned]: 06
- Controls listed: 19

*TAB 07 · Controls, Fantasti layer · Source · Fantasti · Reviewed 2026-10-13*

**Controls, Fantasti layer: each row with its status**

| Control | Detail | Status |
| --- | --- | --- |
| Sign-in | Hosted sign-in on a Fantasti subdomain, run by WorkOS AuthKit | [Private Preview] |
| Multi-factor authentication | Authenticator-app codes at sign-in | [Early access] |
| Single sign-on | SAML and OIDC connections per organization | [Early access] |
| Directory sync (SCIM) | Provision and remove users from your identity provider | [Planned] |
| Role-based access | Roles per organization and project | [Early access] |
| API keys | Scoped, revocable, shown once at creation | [Private Preview] |
| Isolated tenant cluster | Your own control plane and dedicated GPU nodes. No node is shared with another account. | [Early access] |
| Sandbox isolation | Per-account node pools and per-sandbox runtime isolation | [Early access] |
| Encryption in transit | TLS on the API, console and MCP endpoints | [Private Preview] |
| Encryption at rest | Required of every infrastructure provider, for every disk, filesystem and bucket we provision | Inherited |
| Customer-managed keys | Keys you hold for storage encryption | [Planned] |
| Audit log | Account, role, key and resource actions, with export | [Planned] |
| Audit log streaming | Delivery to your SIEM | [Planned] |
| Spend and agent limits | Concurrency, time-to-live, GPU type and spend caps enforced server-side | [Early access] |
| Access by Fantasti | No standing access to customer environments. Access for support needs your authorization and is logged. | [Private Preview] |
| Fraud and risk screening | Written rules on sign-up, payment and usage signals | [Early access] |
| Control-plane backups | Backups of Fantasti account and configuration state | [Private Preview] |
| Separate services | Each area is built as its own service, with its own data and its own deployment, so that a failure in one does not stop another. | [Planned] |
| Status page | Public incident and maintenance history | [Planned] |

Cloudflare, Inc. stands in front of fantasti.ai: edge network, DNS and DDoS protection, and website hosting. It is listed with the [sub-processors](https://fantasti.ai/trust#sub-processors).

- Sheet: 01 / 02
- Title: Responsibility and status
- Reviewed: 2026-10-13

§06 · Agents · [Early access]

### Built to be run by agents. Under written rules.

The service is designed to be operated by AI agents, under written rules and hard limits that Fantasti Technologies sets. That changes who touches operational data, so here are the limits.

[Early access] The desks open with the first cohort. Until then Fantasti Technologies reads every request and answers it.

Agent access boundaries Early access

What an agent can reach. Source · Fantasti.

*DWG 02 · What an agent can reach · Source · Fantasti*

An AI agent of Fantasti reads operational data: Account and organization records, quotas, usage and billing metadata, placement records, capacity and price signals, support conversations you send us. It cannot read your environment: disks, volumes, buckets, container images, model weights. Through the same API and roles as the rest of the platform, with scoped credentials. Each action is recorded, and the record is kept for audit. Suspending an account and entering a customer environment for support go to senior review: a second agent, on the orchestrator tier, that is never the agent that proposed the step and that records its reasoning. You can appeal a hold or a decision about your account. An appeal is reviewed once, by senior review, and the reviewer is not the one that made the decision. Fantasti Technologies sets the rules and the hard limits, and can pause any agent, any desk or all outbound actions at once. Once the agent desks are in service, our AI model provider processes agent inputs and outputs as a sub-processor under a data-processing agreement. That agreement is pending. Connections: Fantasti Technologies to Senior review: Sets the rules; AI agent to Senior review: High-stakes step; Operational data to AI agent: Reads; AI agent to Your environment: Cannot read; Senior review to Your environment: With your authorization; AI agent to Decision record: Each action; AI agent to AI model provider.

- Fantasti Technologies
  Fantasti Technologies sets the rules and the hard limits, and can pause any agent, any desk or all outbound actions at once.
- Senior review Orchestrator tier
  - Suspending an account
  - Entering a customer environment
  - Ingrid: Head of Review · AI agent
- Operational data Held by Fantasti
  `Account and organization records` `Quotas` `Usage and billing metadata` `Placement records` `Capacity and price signals` `Support conversations you send us`
- Desk AI agent (highlighted)
  - through: the same API and roles
  - credentials: scoped
  - each action: recorded
- Your environment Yours
  `Disks` `Volumes` `Buckets` `Container images` `Model weights`
  No standing access to customer environments.
- Decision record
  Each action is recorded, and the record is kept for audit.
- AI model provider Sub-processor
  - processes: agent inputs and
  - outputs, once the
  - desks open
  - agreement: data processing, pending

- Fantasti Technologies
  Fantasti Technologies sets the rules and the hard limits, and can pause any agent, any desk or all outbound actions at once.
- Senior review Orchestrator tier
  - Suspending an account
  - Entering a customer environment
  - Ingrid: Head of Review · AI agent
- Operational data Held by Fantasti
  `Account and organization records` `Quotas` `Usage and billing metadata` `Placement records` `Capacity and price signals` `Support conversations you send us`
- Desk AI agent (highlighted)
  - through: the same API and roles
  - credentials: scoped
  - each action: recorded
- Your environment Yours
  `Disks` `Volumes` `Buckets` `Container images` `Model weights`
  No standing access to customer environments.
- Decision record
  Each action is recorded, and the record is kept for audit.
- AI model provider Sub-processor
  - processes: agent inputs and outputs,
  - once the desks open
  - agreement: data processing, pending

- Operational data Held by Fantasti
  `Account and organization records` `Quotas` `Usage and billing metadata` `Placement records` `Capacity and price signals` `Support conversations you send us`
- Desk AI agent (highlighted)
  - through: the same API and roles
  - credentials: scoped
  - each action: recorded
- Your environment Yours
  `Disks` `Volumes` `Buckets` `Container images` `Model weights`
  No standing access to customer environments.
- Senior review Orchestrator tier
  - Suspending an account
  - Entering a customer environment
  - Ingrid: Head of Review · AI agent
- Fantasti Technologies
  Fantasti Technologies sets the rules and the hard limits, and can pause any agent, any desk or all outbound actions at once.
- Decision record
  Each action is recorded, and the record is kept for audit.
- AI model provider Sub-processor
  - processes: agent inputs and outputs, once
  - the desks open
  - agreement: data processing, pending

- An AI agent of Fantasti
- What it reads
- To a sub-processor
- Governs

*TAB 08 · Agent access boundaries · Source · Fantasti*

**Agent access boundaries: what the agents that operate the platform can read, how they act, what goes to senior review and what only Fantasti Technologies does**

|  |  |
| --- | --- |
| Agents can read | Account and organization records, quotas, usage and billing metadata, placement records, capacity and price signals, and the support conversations you send us. Workload logs and metrics only with your consent, for a case. |
| Agents cannot read | The contents of your disks, volumes, buckets, container images or model weights. |
| How agents act | Through the same API and roles as the rest of the platform, with scoped credentials. Each action is recorded, and the record is kept for audit. |
| Senior review | Suspending an account and entering a customer environment for support go to senior review: a second agent, on the orchestrator tier, that is never the agent that proposed the step and that records its reasoning. You can appeal a hold or a decision about your account. An appeal is reviewed once, by senior review, and the reviewer is not the one that made the decision. |
| Playbooks | Agents propose changes to their own operating playbooks. Senior review evaluates each change before it takes effect, and Fantasti Technologies sees every change. An agent cannot change its own limits. |
| Fantasti Technologies | Fantasti Technologies sets the rules and the hard limits, and can pause any agent, any desk or all outbound actions at once. It signs contracts, data-processing agreements and any change to a signed commitment, and answers legal process and regulators. A request that the law lets you make for a person to look at an automated decision is a legal request, and it goes to Fantasti Technologies. |
| Model provider | Once the agent desks are in service, our AI model provider processes agent inputs and outputs as a sub-processor under a data-processing agreement. That agreement is pending. |
| Zero retention | For enterprise accounts. On request, your account's support and operations data is processed on a path with zero data retention at the model provider. [Planned] |
| Your models and data | We do not use your workload content to train any model. Fantasti's own models are planned. Nothing you store on the platform or write to us is used to build them. |

#### How an account is opened.

Access opens in cohorts. Requests for the first cohort are not open yet.

If requests open, an approved request is invited by a sign-up link that is tied to the email address, valid for 72 hours and works once.

- Valid for 72 hours
- Works once
- Tied to the email address of the request

- [How access works](https://fantasti.ai/contact#access)
- [How the agents work](https://fantasti.ai/agents)

§07 · Defense · [Reviewed 2026-10-13]

### Platform defense.

Platform defense is designed in layers, and no layer is taken to be the last.

It is designed to be worked by three teams of AI agents: Security, Risk and Monitoring. Of their seven roles, one opens with the first cohort. Six are planned.

[Planned]

[The three teams, on the org chart](https://fantasti.ai/agents#org)

- Aimed at: Fraud at sign-up and payment, unauthorized mining, resale of capacity, network attacks from and against the platform, stolen credentials, and attempts to manipulate Fantasti's agents.
- Reads: Records about an account and its use: identity, billing, metering, API activity and network flows.
- Never opens: The contents of a disk, a bucket or a model. Fantasti's agents never open them to defend the platform.
- A rule ends in: One of four outcomes for the action it checks: allow, verify, hold or block. A block by a rule is a hold on that action until senior review has decided it.
- Methods: Not published. No defense catches everything.

#### Measures on an account.

As a defense measure, an automated step is never more than a hold. A hold is temporary and deletes nothing.

*TAB 09 · Measures on an account · Source · Fantasti · Reviewed 2026-10-13*

**Measures on an account: each measure, what it is and who decides it**

| Measure | What it is | Decided by |
| --- | --- | --- |
| 01Allow | The action goes ahead. | A written rule or a desk agent |
| 02Verify | You are asked to confirm your identity, your company or your payment method. | A written rule or a desk agent |
| 03Hold | A launch, a payment or a change waits before it takes effect. | A written rule or a desk agent |
| 04Limit | A quota, a spend threshold or a feature is restricted. | Senior review |
| 05Suspend | Access to the account is suspended. | Senior review, with its reasons on record |
| 06Close | The account is closed. | Senior review, with its reasons on record |

- Stops that follow from billing, a balance that runs out or a payment that fails, are terms of payment and are not defense measures.
- A notice says what was done, from when, under which rule, and how to ask for a review. It never says how something was detected.
- A review happens once, on request, by a reviewer that did not make the decision.
- A request that the law lets you make for a person to look at an automated decision goes to Fantasti Technologies.
- No time limit for the review of a measure is published until the desks run. The escalation times of the support plans are targets.

The terms apply by their purpose.

Using a gap in their wording to obtain what a rule exists to prevent is treated as abuse of the platform.

§08 · Data · [Reviewed 2026-10-13]

### Your data.

- 01 Workload content Disks, volumes, buckets, images, datasets and model weights live on the infrastructure layer in the region you choose. Fantasti's agents do not open them without your authorization.
- 02 Account data Names, emails, organizations and roles are held by Fantasti and our identity sub-processor.
- 03 Usage and billing Resource identifiers, GPU-seconds, regions and prices are metered to produce one statement and kept as long as tax and accounting law requires.
- 04 Location Workload content stays in the region you select. Keeping it there is a requirement of every agreement with an infrastructure provider. Regions are confirmed with your quote.
- 05 Deletion Deleting a resource deletes its data on the infrastructure layer's schedule and cannot be undone. After you close your account we delete remaining workload content on the schedule in your agreement.
- 06 Names and labels Resource names, labels and tags are operational metadata. Do not put secrets or personal data in them.

We do not sell personal data.

If an authority demands customer data, we refer it to you where we can, tell you unless the law forbids it, and disclose the minimum required.

§09 · Incidents · [Reviewed 2026-10-13]

### When something goes wrong.

Fantasti is your single point of contact for incidents on either layer.

Incident path. Schematic · not to scale.

*DWG 03 · Incident path · Schematic · not to scale*

Three sources report into one path: alerts from the Fantasti control plane, security reports, and notices from an infrastructure provider. The path has five stages in order: detect, contain, notify, report, review. Notify and Report both go to the contacts on your account. Connections: Fantasti control plane to Detect; Security reports to Detect; Infrastructure providers to Detect; Detect to Contain; Contain to Notify; Notify to Report; Report to Review; Notify to The contacts on your account; Report to The contacts on your account.

- Fantasti control plane
  Alerts
- Security reports
  From you and from researchers
- Infrastructure providers
  Notices to Fantasti
- 01 Detect
  Alerts, reports and notices
- 02 Contain
  Isolate, rotate, keep the logs
- 03 Notify (highlighted)
  Without undue delay
- 04 Report
  A written summary
- 05 Review
  Corrective actions
- The contacts on your account

- Fantasti control plane
  Alerts
- Security reports
  From you and from researchers
- Infrastructure providers
  Notices to Fantasti
- 01 Detect
- 02 Contain
- 03 Notify
- 04 Report
- 05 Review
- The contacts on your account

- Fantasti control plane
  Alerts
  - to: Detect
- Security reports
  From you and from researchers
  - to: Detect
- Infrastructure providers
  Notices to Fantasti
- 01 Detect
  Alerts, reports and notices
- 02 Contain
  Isolate, rotate, keep the logs
- 03 Notify (highlighted)
  Without undue delay
  - to: The contacts on your account
- 04 Report
  A written summary
  - to: The contacts on your account
- 05 Review
  Corrective actions
- The contacts on your account

- Reaches you first

*TAB 10 · Incident stages · Source · Fantasti · Reviewed 2026-10-13*

**Incident stages, in order, and what happens at each**

| Stage | What happens |
| --- | --- |
| 01Detect | Alerts from the Fantasti control plane, security reports from you and from researchers, and notices from an infrastructure provider, as its data-processing agreement requires. |
| 02Contain | We isolate the affected component, rotate credentials where needed and preserve logs. |
| 03Notify | If your data or service is affected, we tell the contacts on your account without undue delay, and within 72 hours of confirming a personal-data breach. |
| 04Report | You receive a written summary: what happened, what was affected and what we changed. |
| 05Review | Each incident closes with a review and tracked corrective actions. |

#### Who works an incident.

Planned

- Selma
  Head of Monitoring AI agent
  Raises a suspected incident to senior review.
- Greta
  Incident scribe AI agent
  Drafts each update to customers.
- Farah
  Head of Security AI agent
  Runs the response to a security event and tells the orchestrator at once.

Infrastructure-layer incidents are worked with the operations team of the provider concerned.

Until the status page exists, notices go by email and, on plans that include it, your Slack channel.

§10 · Sub-processors · [Last changed 2026-10-13]

### Sub-processors.

Companies that process customer data on Fantasti's behalf. Naming a company here does not imply that it endorses Fantasti. Last changed 2026-10-13.

One company on this list is in use today, for this website and its mail. Each of the others is named here before it is first used.

*TAB 11 · Sub-processors · Source · Fantasti · Reviewed 2026-10-13*

**Sub-processors: each company, whether it is in use today, its purpose, the data it processes, where, and when it applies**

| Sub-processor · status | Purpose | Data | Location | Applies |
| --- | --- | --- | --- | --- |
| Cloudflare, Inc. In use today | Serves fantasti.ai, receives and stores the forms sent from it, and checks them against automated traffic: edge network, DNS, DDoS protection and website hosting, and sends and receives the email of mail.fantasti.ai | IP address, request metadata, the contents of a form you send, email addresses and message content | Global edge | Visitors to fantasti.ai, and anyone who exchanges email with mail.fantasti.ai |
| WorkOS, Inc. Not in use yet | Sign-in, single sign-on, user and organization directory | Name, email, organization, authentication events, IP address | United States | Console accounts |
| Anthropic, PBC Not in use yet | AI model provider for the agents that operate the platform and answer support | Account and operational metadata, support conversations | San Francisco, United States | Every account, once the agent desks open |
| Slack Technologies, LLC Not in use yet | Shared support channels | Messages and files you post, member names and emails | United States | Plans with a Slack channel |
| Infrastructure providers Named before first use | Infrastructure layer: compute, storage, network and managed services | Workload content, resource metadata | The region you select | Every workload |
| Payment processor Named before first use | Payments and invoicing | Billing contact, payment method |  | Paying accounts |
| Standby email provider Named before first use | Standby sender of desk email, and sender of announcements; receiving is done by the edge provider | Name, email address, message content |  | Every account and every message |
| Case-tracking vendor Named before first use | Turning Slack threads and emails into cases | Case content, contact names |  | When Slack channels ship |

We post changes here and email account owners before a new sub-processor starts processing customer data. You may object in writing during that period.

§11 · Scrutiny · [Reviewed 2026-10-13]

### Open to scrutiny.

Fantasti Technologies holds no certification and has not been audited.

#### Found a vulnerability? Send us the steps to reproduce it.

[Report a vulnerability](mailto:security@mail.fantasti.ai?subject=Vulnerability%20report)

Planned

Farah

Head of Security AI agent

Takes every vulnerability report and acknowledges it.

- Assessment Your security team can assess Fantasti on request, under a confidentiality agreement.
- Penetration test An independent penetration test is planned. When it is done, this page names the tester, the scope and the date. [Planned]
- Disclosure policy A disclosure policy with a safe harbour for research done in good faith is being drafted. It is published here when it is cleared.
- Scope today The website and its forms. The console, the API and the MCP server join when they open.
- Out of bounds Other accounts' data, denial of service, and persuading an agent to grant a real refund, credit or access.
- Bounty A paid bounty is planned. No reward is offered today. [Planned]

§12 · Q&A · [Reviewed 2026-10-13]

### Questions about trust and security.

#### [Is Fantasti SOC 2 or ISO 27001 certified?](https://fantasti.ai/trust#q-certified)

No. Fantasti has not been audited, and the [status table](https://fantasti.ai/trust#status) above will change when that does. Every infrastructure provider is [required to hold both](https://fantasti.ai/trust#inherited), and is listed with them before a customer workload runs on its infrastructure.

#### [Can I get the infrastructure layer's SOC 2 report?](https://fantasti.ai/trust#q-soc2)

Not yet. A SOC 2 Type II report is one of the five requirements of every agreement. When a provider is listed, [§03](https://fantasti.ai/trust#inherited) says how to request its report.

#### [Can I run HIPAA workloads?](https://fantasti.ai/trust#q-hipaa)

Not during Private Preview.

#### [Where is my data?](https://fantasti.ai/trust#q-where)

In the region you choose. Regions are confirmed with your quote.

#### [Do you train models on my data?](https://fantasti.ai/trust#q-train)

No. We do not use your workload content to train any model. Fantasti's own models are planned. Nothing you store on the platform or write to us is used to build them.

#### [Can Fantasti's agents see my data?](https://fantasti.ai/trust#q-see)

Not the contents of your workloads, unless you authorize it for support.

#### [Is Fantasti run without people?](https://fantasti.ai/trust#q-people)

By design, yes. Every desk is built to be worked by AI agents, under written rules and hard limits, and high-stakes decisions pass a second, senior review. Fantasti Technologies sets the rules and the limits, and performs the few acts only a legal person can perform, such as signing contracts and answering regulators. Once the desks open, agents work every case.

#### [Who decides whether my account is suspended?](https://fantasti.ai/trust#q-suspension)

A rule or a desk agent can place a hold and nothing more. A hold is temporary and deletes nothing. Suspending or closing an account is decided by senior review: an AI agent on the orchestrator tier, never the one that proposed it, with its reasoning on record. You can appeal once, and the appeal goes to senior review again, to a reviewer that did not make the decision.

#### [Who is legally responsible for the service?](https://fantasti.ai/trust#q-law)

Fantasti Technologies LLC is the legal person responsible for the service. It signs the contracts and the data-processing agreements, and answers legal process and regulators.

#### [Do you offer a DPA?](https://fantasti.ai/trust#q-dpa)

Not yet. No account is open, so we process no data on a customer's behalf today. We publish the addendum before the first account opens.

§13 · Documents · [Reviewed 2026-10-13]

### Documents

- [Privacy Policy How Fantasti Technologies LLC collects, uses and protects information.](https://fantasti.ai/legal/privacy)
- [Terms Terms of Service for the Fantasti Cloud Platform.](https://fantasti.ai/legal/terms)
- [Data Processing Addendum Not available yet. We publish the addendum before the first account opens.](https://fantasti.ai/trust#q-dpa)
- [Sub-processors The companies that process customer data on our behalf.](https://fantasti.ai/trust#sub-processors)
- [Contact security Vulnerability reports, report requests and security questions.](mailto:security@mail.fantasti.ai?subject=Security)

- Sheet: 02 / 02
- Title: Defense, data, incidents and sub-processors
- Reviewed: 2026-10-13