AI Disclaimer

Claycable is a platform for training, deploying, and chatting with AI agents. Everything below is a plain-English description of how we expect you to think about outputs from those agents and the fine-tuned models behind them.

1. Outputs are model-generated

Every response you see in a Claycable agent, every email it drafts, every tool call it makes, and every artifact it generates is produced by a language model — either a base model we host or a fine-tuned adapter you (or another user) trained. Models are statistical systems. They are very good at sounding confident and very capable of being completely wrong.

2. Not professional advice

Outputs from Claycable agents are not legal advice, medical advice, financial advice, tax advice, mental health advice, safety advice, or advice of any kind that you would ordinarily expect from a licensed professional. If a Claycable agent tells you something that would materially affect your health, money, freedom, or wellbeing, treat it as a starting point for conversation with a real human professional — not as the final word.

3. Verify before acting

You are responsible for verifying outputs before relying on them. This applies especially to:

• Numbers, dates, names, citations, and quotes — models hallucinate these regularly
• Code, queries, configuration, and anything that will be executed
• Anything an agent says about its own capabilities, limits, or memory
• Anything an agent says about Claycable's features, pricing, or policies (check the docs)

4. Fine-tuned model ownership

When you fine-tune an adapter on Claycable, the resulting weights are associated with your account and you retain ownership of them, subject to the Terms of Service. Outputs generated by your fine-tuned models are your responsibility, including ensuring they comply with applicable laws and third-party rights. We do not screen, moderate, or guarantee the outputs of user-trained adapters.

5. Training data and provenance

You are responsible for the data you upload for fine-tuning. If a fine-tuned model produces outputs that infringe a third party's rights because it was trained on material you did not have the right to use, that is your liability, not ours. We do not audit training datasets for copyright, consent, or licensing.

6. Agents acting on your behalf

Claycable agents can be given tools that send messages, write to memory, call webhooks, open HTTP requests, and (eventually) open pull requests against code repositories. When you grant an agent a tool, you are authorising it to take real actions on your behalf. Side effects are your responsibility. Review tool calls carefully before granting a new agent permissions, and use the deploy-token / approval workflows we provide.

7. Model and infrastructure changes

We may change the underlying base models, inference providers, or infrastructure used by the platform. These changes can shift the behavior of any agent, including agents you have already built and deployed. We will give reasonable notice for material changes, but small drifts in response style or tool-call patterns are normal and not actionable bugs.

8. Bias, harm, and edge cases

Models reflect biases in their training data. Despite safety tuning, base models occasionally produce outputs that are offensive, factually wrong, or unsafe in edge cases. If you encounter outputs you believe are harmful, please report them to safety@claycable.com. Operators of public-facing agents are responsible for their own safety review.

9. No warranty of fitness for purpose

Nothing on this page or anywhere else on Claycable should be read as a warranty that any specific model, adapter, or agent is fit for any specific use case. You are the operator. You decide what to trust and what to deploy.

10. Contact

Questions about this disclaimer go to legal@claycable.com. Safety reports go to safety@claycable.com.