AI and Model Disclosures

1.1. The Platform is an AI system. When you send a prompt through the Web Interface, Control Panel, or an automated agent configuration, the Platform generates outputs — text, code, plans, and other content — using artificial-intelligence models. You are interacting with an AI, not with a human.

1.2. The Platform is built around a managed Agent Instance running an open-source agent runtime. The agent plans tasks, uses tools, executes code in sandboxes, and maintains memory, producing AI-generated outputs.

1.3. Cyphra operates the Platform and the agent runtime. Unless you supply your own model keys (Section 3.2), the underlying machine-learning models are operated by third-party model providers, as described below.

2.1. AI outputs may be wrong. AI-generated outputs may be inaccurate, incomplete, biased, offensive, or outdated. Models can "hallucinate" — produce confident-sounding but false statements. Outputs reflect patterns in training data, not verified truth, and may not reflect events after a model's training cutoff.

2.2. Not professional advice. Outputs are not legal, medical, financial, tax, or other professional advice, and do not create a professional relationship of any kind. Always consult a qualified professional before acting on AI output in these areas.

2.3. Human review recommended. You should have a qualified human review any AI output before relying on it for consequential decisions — including decisions affecting legal rights, health, safety, finances, employment, or third parties.

2.4. No fitness guarantee. Outputs are provided "as is". We do not guarantee that outputs will be fit for any particular purpose, meet your expectations, or be uninterrupted, error-free, or available at any specific quality level. Capability and behavior may change as models and configurations change.

2.5. Model behavior. Each model provider sets its own safety behavior, filters, and content policies. Identical prompts may produce different outputs through different providers or at different times.

3.1. Cyphra-provided model access. Where you use model access provided by Cyphra, your prompts are routed to third-party model providers selected by us. Those providers process prompts and outputs under their own terms, usage policies, and privacy practices, which apply to that processing. The current providers and their policies are listed in our Subprocessor List, which we maintain and update.

3.2. Bring your own keys (BYOK). You may alternatively connect your own API keys for model providers. In that case:

• (a) your relationship is directly with that provider, governed by the provider's own terms and usage policies, with which you are responsible for complying;
• (b) your keys are encrypted at rest and never logged by us, and remain your property at all times; and
• (c) provider-side usage, quotas, costs, and policy compliance are your responsibility.

3.3. Changes. We may change the model providers used for Cyphra-provided access. Material changes affecting processing of prompts and outputs will be reflected in the Subprocessor List in accordance with our Terms & Conditions and Data Processing Addendum.

4.1. Cyphra does not train models on Customer Content. We do not use your prompts, uploads, agent memory or configuration, or AI outputs to train, fine-tune, or improve any machine-learning model, whether ours or a third party's.

4.2. Upstream provider policies. When you use Cyphra-provided model access, some upstream model providers may, under their own terms and policies, use interactions with their models for purposes such as improving their services. That use is governed by the provider's own policy, not by us. Current provider policies are listed in our Subprocessor List — review them before sending content you wish to keep out of provider-side processing.

4.3. BYOK alternative. If provider-side handling is unacceptable to you, using your own keys (Section 3.2) with a provider whose terms meet your requirements may give you more direct control of that relationship.

5.1. Prompts and outputs transit through your Agent Instance and — for the selected model — to the relevant model provider and back. Any tool, integration, or external endpoint your agent configuration accesses may also receive data you send to it.

5.2. Retention in your instance. Prompts, uploads, agent memory and configuration, and AI outputs are stored in your Agent Instance under your control, subject to our Terms & Conditions and Data Processing Addendum. You can view, manage, and export Customer Content through the Web Interface where features allow.

5.3. No card data. Billing is handled through the Billing Portal; card data is processed by the third-party payment gateway at checkout and never touches Cyphra systems.

6.1. You are interacting with an AI system. The Platform is an AI system, and its outputs are AI-generated. We present the service transparently as AI-driven, and users are informed that they are interacting with AI.

6.2. Jurisdictional posture. Cyphra operates from New York / New Jersey, USA, and US law is the primary legal framework for the Platform. AI-transparency obligations under non-US regimes apply where required for users located in the EU/UK — for example, under the EU AI Act, which imposes transparency and other obligations on providers and deployers of AI systems, including disclosure where users interact with AI and labeling of certain AI-generated content. This section is a generic, good-faith statement of transparency and does not constitute our full AI Act compliance analysis.

7.1. Customer oversight. You are responsible for oversight of the AI and review of its outputs in your use of the Platform, including reviewing outputs before acting on them or publishing them.

7.2. Automated agent actions. The agent runtime can be configured to take actions autonomously (for example, running code, calling tools, or contacting external services). If you configure agents to take actions, you are the party deploying that configuration, and those actions and decisions are yours. Cyphra provides the runtime but does not make, approve, or own the decisions your configured agents carry out.

7.3. Role allocation. The division of responsibilities between Cyphra, the customer, and upstream model providers under the EU AI Act (where users are located in the EU/UK) and under US law and similar regimes — in particular who is "provider" and who is "deployer" for the components involved — is a legal allocation question.

8.1. Your use of the AI features — including the prompts you send and the outputs you act on — is subject to our Acceptable Use Policy. That policy prohibits unlawful, harmful, infringing, deceptive, and abusive uses, and applies in full to AI-generated content.

9.1. AI outputs may not be protectable. Under current United States Copyright Office guidance, material generated purely by AI without sufficient human creative input is generally not protectable by copyright, and protection may be limited or unavailable for AI-generated content in other jurisdictions as well. We make no representation that you hold any IP rights in outputs.

9.2. Your responsibility to verify rights. Before using any output commercially, you are responsible for verifying that you have the rights to use it — including checking for third-party rights in the output (for example, reproduced text, code, or images from training data), assessing the protectability of the content in your target markets, and confirming compliance with the terms of the applicable model provider. Outputs that reproduce or closely imitate third-party material may itself constitute infringing content under Section 2.3 of our Acceptable Use Policy, and using the Services to generate or distribute such content is prohibited by that policy.

9.3. Ownership of Customer Content, including AI outputs stored in your instance, is governed by our Terms & Conditions. Nothing in this section grants you rights in third-party material appearing in outputs, or in the underlying models.

10.1. You must not use the Platform to generate deceptive impersonation content — including deepfakes, voice or likeness clones, or fabricated content presented as real — or content that impersonates any real person, brand, or organization in a manner that deceives or is intended to deceive. This is a prohibited use under Sections 2.6 (impersonation and deceptive practices) and 3.8 (false reports) of our Acceptable Use Policy, and may also constitute harassment or doxxing under Section 2.5 of that policy where directed at an identified person.

10.2. Synthetic or altered media of real persons must not be created or shared without clear and conspicuous disclosure that the media is synthetic or altered and, where required, the consent of the person depicted. This applies regardless of intent and includes media that is lawful in your jurisdiction. See our Acceptable Use Policy for the full rules on deceptive and harassing content.

10.3. Applicable synthetic-media laws. Several US states — including New York (civil liability for the nonconsensual commercial use of a person's likeness, including digital replicas) and New Jersey (criminal prohibitions on certain deceptive media and impersonation) — and, where users are located in the EU/UK, EU regimes impose disclosure, consent, or civil/criminal liability rules for synthetic media, political deepfakes, and nonconsensual likeness use.

• Legal and AI disclosure questions: [email protected]
• Privacy and data-protection questions: [email protected]
• General support: [email protected]

This document describes the Platform's AI features. It is provided for information and does not modify our Terms & Conditions, the Acceptable Use Policy, or the Data Processing Addendum.