AI Transparency

Effective Date: July 27, 2026

Read this first:Raileon agents are AI systems, not people. They do not have a name, a desk, or a job title, even when a message reads as though they do. Anything an agent writes, summarises, or calculates is machine-generated output. A human being at the business using Raileon — not the agent, and not Raileon LLC — is accountable for that output and for anything done with it.

This page is written for two audiences. If you are a Raileon customer, it tells you what our agents do, where they fail, and what we expect of you. If you received an email, message, or document that was produced with the help of a Raileon agent and you never agreed to anything with us, it tells you what you are looking at and how to raise a problem.

1. What Our Agents Are

Raileon LLC is a Florida company that builds and operates AI agents for businesses — mostly professional-services firms such as law practices, accounting firms, and consultancies. An agent is software. It runs on a dedicated instance we operate for the customer, and it is configured by that customer to perform specific work.

Depending on how a customer configures it, an agent may:

Read and triage incoming email and messages.
Draft replies, follow-ups, notes, and other written material.
Summarise documents, transcripts, and long threads.
Extract structured data — names, dates, amounts, deadlines — from unstructured text.
Look things up in the tools the customer has connected, such as their mailbox, calendar, or CRM.
Take scheduled or triggered actions, such as sending a reminder or posting to a chat channel.

An agent has no independent authority. It can only touch the systems a customer connects and can only do the things a customer switches on.

2. What Is Machine-Generated

When an agent is involved, assume the following are produced by an AI system rather than typed by a person:

Drafted emails and messages— including replies sent from a customer’s address or chat account, whether or not a person edited them first.
Summaries— of calls, threads, meetings, contracts, or files.
Reports and analyses— including any figures, totals, or trends they contain.
Extracted data— fields pulled out of a document and written into another system.
Scheduled and automated actions— reminders, follow-up sequences, status updates, and similar routine steps.

A person may review and edit any of these before they go out. Whether that happened in a particular case is a decision made by the customer, not by us.

3. Known Limitations

AI systems are useful and they are also unreliable in specific, predictable ways. We would rather say this plainly than bury it.

They can be confidently wrong. A generated answer can be fluent, well-structured, internally consistent, and still false. An agent will not usually signal that it is unsure. Confidence in the writing is not evidence that the content is correct.
They misread context.Sarcasm, hedging, an exception buried in a long thread, an unstated assumption inside a firm, or a detail that matters only to one client — all of these get missed or flattened.
They go stale. Model knowledge has a cut-off, and an agent only sees the data it has been given access to. Prices, rules, statuses, and contact details can be out of date without anything looking wrong.
They are inconsistent. The same question asked twice can produce different answers.
They invent specifics. Citations, case names, section numbers, dates, and figures are exactly the sort of detail an AI system will fabricate to complete a plausible-sounding sentence.

The practical rule: material facts, figures, dates, deadlines, and anything legal, financial, medical, or contractual must be checked by a person against the source before anyone acts on it. Do not treat agent output as verified simply because it came out of the platform.

4. Human Accountability

The customer is responsible for reviewing and approving consequential actions. This is not a disclaimer we hide in the terms — it is how the product is built.

Approval gates. Actions can be held for a human to approve before they execute. Sending an email, posting a message, and similar outward-facing steps can all be gated this way.
Autonomy controls. Customers choose how much an agent may do on its own, per capability, and can narrow that at any time.
Activity logs. Agent actions are recorded so a person can see what happened and when.

We strongly recommend leaving approval gates on for anything that leaves the business, commits the business, moves money, or affects a legal deadline. If a customer turns those gates off, the output still goes out under their name and remains their responsibility.

5. Telling Your Own Clients

This section is for Raileon customers. If your agent contacts the people you serve, they deserve to know when they are reading something an AI system produced.

What we expect of you

Do not present an AI-generated message as something you personally wrote. Do not give an agent a human persona, a fabricated staff name, or a signature that implies a person composed the message.
Where an agent is handling a conversation with little or no human review, say so — a single plain line is enough, for example that the message was drafted with AI assistance and is reviewed by your team.
If someone asks whether they are talking to a person, answer honestly and promptly.
Follow whatever your own professional body, regulator, or client engagement letter requires. Those obligations sit with you, and they are often stricter than ours.

Why this is increasingly a legal requirement

Disclosure is turning from good manners into law. Under Article 50 of the EU AI Act, providers and deployers of systems that interact with people generally must ensure those people are informed that they are dealing with an AI system, and certain AI-generated content must be marked as such. A growing set of US state laws impose their own AI-disclosure duties, particularly around chatbots, marketing, and automated outreach, and the list grows with each legislative session.

We are describing the general direction of the law, not giving you legal advice. Whether a specific rule applies to your firm and your outreach is a question for your own counsel. See our Acceptable Use Policy for the conduct rules that apply to your use of the platform.

6. Models and Data

Agent reasoning runs on large language models supplied by third-party AI providers. We access those models through commercial APIs under contract, with model training disabled on the provider side.

Customer data is sent to a third-party AI provider only to produce the requested output, and is not used to train public AI models.
Each customer runs on a dedicated, isolated instance. Data is encrypted in transit with TLS 1.2 or higher and at rest with AES-256.
All processing takes place in the United States.
The full list of the providers we rely on, and what each one does, is published at raileon.com/subprocessors.

For how we collect, store, and delete information generally, see our Privacy Policy.

7. Not Professional Advice

Nothing an agent produces is legal, medical, tax, accounting, or financial advice, and no agent output creates a professional relationship of any kind with Raileon LLC. Our customers are often licensed professionals; when they use an agent to help draft something, the professional judgement, the review, and the advice are theirs. The agent is a drafting tool, in the same category as a word processor with opinions.

8. Automated Decisions About People

Raileon does not use agents to make solely-automated decisions that produce legal effects concerning individuals or that similarly significantly affect them. We do not screen, score, rank, or reject people by algorithm.

Customers must not either. Using an agent to decide, without meaningful human involvement, whether someone gets a job, credit, housing, insurance, healthcare, education, or a legal outcome is prohibited by our Acceptable Use Policy. An agent may help gather and summarise information for a person who makes the decision; it may not be the one making it.

Regardless of where you live, we extend the same data-protection rights to everyone — including the right not to be subject to a decision based solely on automated processing. We do not gate those rights by country. To exercise them, contact privacy@raileon.com.

9. Reporting a Problem

If you received a message you believe was AI-generated and something is wrong with it — it is inaccurate, misleading, unwanted, impersonates someone, or was not disclosed as AI-assisted — tell us. This applies whether you are a customer or someone a customer contacted.

Email admin@raileon.com.
Include the message if you can, or a screenshot, plus roughly when you received it and who it appeared to come from. That is what lets us trace it back to the account responsible.
For a suspected security issue or data exposure, write to security@raileon.com instead.
For data-protection requests — access, correction, deletion — write to privacy@raileon.com.

We will acknowledge reports about agent-generated messages and investigate. Where a customer has broken our rules, we act on the account. We cannot control what a customer writes with an agent any more than an email provider controls what is typed into an email, but we can and do enforce our policies against accounts that misuse the platform.

10. Changes to This Statement

We will update this page as the technology and the law change. The Effective Date at the top shows when it was last revised. Questions about anything here can go to admin@raileon.com.

Raileon LLC
Miami, Florida