AI Policy
How Crypton uses artificial intelligence on our platform.
Last updated: September 2026
Overview
Crypton uses artificial intelligence in two distinct ways. Platform AI runs on self-hosted infrastructure we control and is not optional for SMS abuse prevention. Crypton AI (Chat and the API) is an optional product you turn on: prompts go to Seeweb S.r.l. in Italy, Inco AI, Inc. in the United States, or Abliteration AI, Inc. in the United States, depending on the model you pick, and are not retained as prompt bodies after the reply. We do not use OpenAI, Google, or Anthropic. We still do not use AI for customer support.
For broader data handling practices, see our Privacy Policy. PRIVACY POLICY
Outgoing SMS Content Filter
Outgoing SMS messages sent through Crypton phone numbers are screened by an AI content filter before delivery.
Purpose
- Detect and block spam, phishing, and fraud in outgoing messages
- Protect recipients and maintain platform integrity
- Applies to outgoing messages only — incoming messages are not screened by AI
How It Works
- Messages are processed in real time by our self-hosted custom model running on Crypton infrastructure
- Only the message text is sent to the filter — no metadata beyond what is needed for screening
- Short verification codes (alphanumeric strings without spaces) bypass the filter automatically
- If the filter is temporarily unavailable, messages remain pending until screening can resume — they are not sent unscreened
Data Handling
- Messages that pass screening are not retained by the AI filter
- Blocked messages are encrypted at rest and retained so you can see them in your conversation, and so our team can review them if you appeal
- You are not charged for blocked messages
- No third parties are involved in the SMS screening process
Internal Content Translation
Crypton uses a self-hosted TranslateGemma model to translate Crypton-authored content into other languages.
What Is Translated
- Blog posts and articles published on crypton.sh
- Site-wide notifications and announcements
- Product roadmap feature descriptions
What Is Not Translated
- Your SMS messages or conversation content
- Personal account data or billing information
- Any user-generated content outside of Crypton-authored pages
Optional Crypton AI (Chat and API)
Crypton Chat and the OpenAI-compatible Crypton AI API are optional. Guest Chat uses the free model without an account. The API and paid models need an account. Nothing is sent to the inference provider until you send a Chat message or call the API.
How inference works
- Crypton is the subscriber. We issue our own Chat session tokens and API keys. Signed-in users authenticate with Crypton credentials. Guest Chat is capped and is not saved.
- EU models are proxied to Seeweb S.r.l. on EU GPUs in Italy. Default retention at that provider is zero. A third-party zero-data-retention certificate is dated 24 July 2026.
- US models are proxied to Inco AI, Inc. Their published policy is zero retention of request and response bodies (not a third-party certificate). They keep usage metadata for their own billing.
- Unrestricted US models are proxied to Abliteration AI, Inc. Their published policy is zero retention of request and response bodies (not a third-party certificate). Those models are unrestricted; Crypton’s own abuse rules still apply. They keep usage metadata for their own billing.
- When Search is Auto or On, that query goes to Crypton’s SearXNG host, which asks DuckDuckGo, Wikipedia, and Wikidata. Crypton may fetch the public page: its text, and on vision models a few images from the page. The inference provider still does not keep the chat.
- Crypton does not log prompt or completion bodies on the gateway. We store usage metadata only (model, token counts, cost, timestamps) for billing and abuse prevention.
- Chat history is a separate feature: after the reply, we may store Halite-encrypted ciphertext on Crypton until you delete the thread. You can start a chat that is not saved. Encrypted history is not the same as inference retention.
- You choose the lane when you choose the model. Crypton is a UK company (Rinzler Labs, Scotland). Billing metadata is retained under UK law.
Acceptable use
- You must not use Crypton AI for unlawful content. Unrestricted models still sit under Crypton’s own abuse rules. We pass through the upstream acceptable-use rules.
- We may suspend keys or Chat if we detect abuse, including attempts to exhaust the free allowance or bypass billing, as described under Suspension and Termination in our Terms.
What We Do Not Do
- We do not train AI models on your messages or personal data
- We do not use AI chatbots for customer support
- We do not send SMS, email, or account data to external AI APIs. Optional Crypton AI is the only user-initiated exception, and only Seeweb, Inco, or Abliteration is used for that product
- We do not use AI to read, analyse, or monitor incoming SMS messages
- We do not use AI for advertising, profiling, or behavioural tracking
Your Choices & Appeals
If your outgoing message is blocked by the content filter, you can revise and resend it. Blocked messages appear in your conversation so you know what was held back. You can delete Crypton Chat threads at any time. You can revoke API keys from your dashboard.
If you believe a message was blocked in error, open a support ticket from your account or reach us via SimpleX Chat to appeal. Our team will review the block and respond as quickly as possible.
Contact
Questions about this policy or how Crypton uses AI:
The English version of this policy is authoritative. Translations are provided for convenience.