01Purpose of This Disclosure
This document discloses the material risks of using AI-powered products and services provided by Prathos AI. We publish it because AI is a new technology category, the failure modes are not always obvious, and we want clients and end-users to make informed decisions. Read this before going live with any Prathos-built system.
02AI Accuracy and Hallucination Risk
Large language models — the technology underneath our chatbot products — generate responses by predicting plausible next text, not by retrieving verified facts. They can and do produce confidently-stated information that is incorrect, outdated, or entirely fabricated. This is called "hallucination" and there is no current technique that eliminates it.
We mitigate this by:
- Constraining the model to a knowledge base you define (so it cites your information, not its training data)
- Configuring fallback behavior — when the model is uncertain, it says so and offers to connect the user to a human
- Logging every conversation for review and refinement
Mitigation reduces risk. It does not eliminate it. Operators must monitor conversation logs and correct misstatements promptly.
03Business Outcomes Are Not Guaranteed
We share case-study metrics on our marketing pages (booking lift, lead-volume growth, revenue impact). These are real results from specific engagements with specific operators in specific markets. They are not predictions of your results. Outcomes depend on factors outside our control: your offer, your traffic, your follow-through, your market, your team. We do not guarantee any specific business outcome.
04Data Handling by AI Providers
By default, our chatbot product sends conversation content to a third-party large-language-model provider (such as OpenAI or Anthropic) to generate responses. This means:
- The provider sees the user's message and the relevant knowledge-base context we send with it
- Provider policies govern what they do with that data — we use providers with no-training commitments where available, but those commitments can change
- If you handle regulated data (PHI, financial account information, etc.), discuss self-hosted model options before deploying
See our Privacy Policy for the full data-flow description.
05Third-Party Service Dependencies
Our services depend on third-party platforms: AI model providers, hosting infrastructure, payment processors, integration partners (Google, HubSpot, Cal.com, n8n, and others). Outages, API changes, or policy shifts at those providers may affect availability or behavior of services we deliver. We monitor for these and respond as quickly as practical, but we do not control them.
06Compliance and Regulatory Limitations
Standard Prathos deployments are not, by default:
- HIPAA-compliant — covered entities (healthcare providers) need a Business Associate Agreement and additional controls before processing PHI through our systems
- PCI-DSS compliant — do not pass card numbers through our chatbots
- SOC-2 certified — we have not undergone a formal SOC-2 audit
For regulated industries, we offer custom configurations: self-hosted open-source models, on-your-infrastructure deployment, BAAs where applicable. Discuss your compliance requirements before scope.
07Operational and Uptime Risks
We target high availability but do not currently offer a formal SLA on productized plans (Spark, Growth, Scale). Custom enterprise engagements include negotiated SLAs. Past incidents — including model-provider outages — have resulted in temporary degraded responses (the bot reverts to a "let me have someone follow up" mode) rather than complete failure. Plan for the possibility of brief outages.
08Recommended Mitigations
If you are deploying a Prathos system in your business, we recommend:
- Review the daily CSV digest each morning during the first 90 days
- Configure escalation paths so high-value or sensitive intents route to a human
- Avoid AI-stated commitments on price, availability, or policy unless those values are confirmed against a live source
- Maintain human review for any output sent to customers via email, SMS, or other persistent channels
- Re-train the knowledge base when your business changes (new services, new pricing, new providers)
09Acknowledgment and Contact
By using Prathos AI services, you acknowledge that you have read and understood this Risk Disclosure. This document is intended to inform, not to replace independent professional advice. For regulated industries, consult counsel and compliance specialists before deployment.
Questions: ryan@prathos.com.