AI Use Policy

The short version#

• AI output can be wrong even when it sounds certain. Check anything that matters.
• You are responsible for what you do with an Output, including publishing it, running it or acting on it.
• Ethen is not a lawyer, doctor, accountant or engineer. Use qualified people for decisions in those areas.
• Different models behave differently, and the same model can answer the same question differently on different occasions.
• A model suggesting an action is not Ethen taking it. When Ethen can act for you, you approve consequential actions first.

Why AI output needs checking#

Ethen's answers come from large language models and media models operated by Model Providers. These models predict likely text, images or sounds. They do not look facts up the way a database does, and they can:

• state things that are false, including invented facts, quotations, statistics, case citations and references (often called hallucinations);
• be out of date, because a model's knowledge stops at a point in time unless web search supplies newer information;
• misunderstand your request, especially when it is ambiguous or relies on context the model does not have;
• reflect biases present in the data they were trained on;
• give different answers to the same question, or change their answer when challenged, whether or not the original was correct;
• sound equally confident whether they are right or wrong.

How to verify output#

Match the effort to the stakes. A brainstorm needs little checking; a contract clause, a dosage, a security fix or a financial figure needs a lot. Good practice includes:

• checking facts, names, numbers and dates against a primary source;
• opening the sources Ethen cites and confirming they say what the answer claims;
• asking for reasoning or alternatives, and testing the answer against what you already know;
• running and testing generated code before relying on it;
• having a qualified person review anything with legal, medical, financial or safety consequences;
• keeping a record of how you checked important work, especially in professional settings.

Research, search and sources#

When you turn on web search or deep research in Ethen Chat, Ethen searches the web, reads pages and gives an answer with citations. Citations make checking easier, but they are not proof:

• a model can cite a real page for a claim the page does not make;
• it can summarize a page inaccurately or miss important context;
• sources themselves can be wrong, biased, outdated or paywalled;
• search coverage is limited, and an answer may rest on a small number of pages.

Read the cited sources for anything you will rely on. Ethen does not guarantee that research results are complete or that every relevant source was found.

Professional and high-stakes use#

Do not treat Ethen as a substitute for professional advice. Use it to prepare, draft and explore, and have a qualified professional review the result, when the work involves:

• legal matters: contracts, legal rights, litigation, regulatory obligations;
• health: diagnosis, treatment, medication, mental-health crises;
• finance: investment decisions, tax positions, lending, accounting judgments;
• safety-critical engineering: structural, electrical, medical-device or industrial-control systems;
• decisions about people: hiring, firing, credit, housing, insurance, education, benefits, immigration or law enforcement.

For decisions about people, the Acceptable Use Policy requires human review and compliance with the laws that apply. Many jurisdictions regulate automated decision-making, and you are responsible for meeting those rules in your own use.

Generated code#

AI-generated code can look correct and still be wrong. Before you rely on it:

• read it and make sure you understand what it does;
• run it in a safe environment and test it, including edge cases and failure paths;
• check it for security problems, such as injection flaws, unsafe handling of secrets, overly broad permissions or vulnerable dependencies;
• check that suggested dependencies exist, are maintained and are what they claim to be, because models sometimes suggest packages that do not exist;
• confirm you have the rights to use any code that may come from a licensed source;
• never paste production secrets, private keys or credentials into a Prompt unless the feature is designed to handle them securely.

When Ethen Code becomes available, it will propose changes and ask before taking actions such as pushing to a repository. Approving a change means you have reviewed it. See Ethen Code & Repository Data.

Security findings#

If you use Ethen to review code or systems for vulnerabilities, treat its findings as leads, not conclusions. Models can report problems that do not exist and miss ones that do. Confirm each finding, assess its real impact in your environment, and follow your organization's disclosure and remediation process. Keep security testing within the scope of your authorization, as the Acceptable Use Policy requires.

Creative work and synthetic media#

Ethen Studio and other features can generate images, video, audio and voices. When you use generated media:

• make sure you have the rights and consents needed for any real person, brand or work it depicts or imitates (see Ethen Studio Media, Likeness & Generated Content);
• do not present synthetic media of real people as authentic;
• consider disclosing that content is AI-generated where your audience would reasonably want to know, and follow any disclosure rules that apply to you, such as advertising or election rules;
• remember that generated content may resemble existing work, and that similar outputs may be generated for others.

Model and provider differences#

Ethen gives you access to models from several Model Providers. By default, Ethen Chat's conversational experience, Faros, currently runs on Meta's Muse model. Other modes and models you choose use other providers. Models differ in accuracy, reasoning style, knowledge cut-off, safety behavior and how they handle images or long documents.

Providers also update and retire models, which can change how Ethen responds over time. If a result matters, note which model produced it, and re-check results when you change models. The Privacy Policy explains which providers receive your Prompts.

Actions, automation and approvals#

There is a difference between Ethen suggesting something and Ethen doing it:

• Suggestions stay within Ethen. If a model says "I'll send that email", nothing is sent unless a feature built for that action carries it out with your authorization.
• Actions happen in other services you have connected, such as creating an email draft, adding a calendar event or sending a message.

When connected apps become available, Ethen will run read-only lookups in connected services as part of your request. It will ask for your approval before creating drafts, changing events, or sending or posting messages. You can grant time-limited standing approval for some lower-risk actions; sending and posting always require approval each time. Before approving, check:

• the account the action will use;
• the recipients;
• the exact content;
• whether the action can be undone.

Connected Apps & Integrations lists each action and its approval rule.

Ethen does not currently run unattended automations. If it does in the future, think carefully about what an automation can do without you, and limit it to what you would approve each time.

Local and private processing#

Some announced Ethen features run on your own computer, such as Ethen Code's local runtime and local model support. Local processing can keep certain data on your device, but it does not make Ethen fully offline or private by definition. Your Account, sign-in, settings and any cloud model you choose still use the network, and local models may be less capable than cloud models. Check each feature's documentation for which parts run locally.

Connected data#

When connected apps become available, content from your email, files or calendar may be included in what a model sees when answering you. Information from connected sources can be inaccurate, out of date or written by someone else. Instructions hidden in a document or email can try to steer a model ("prompt injection"). Be cautious when a model acting on connected data proposes an action, and check the details before you approve it.

Your responsibility#

Ethen is a tool. You decide what to ask, which Outputs to rely on, what to publish and which actions to approve, and you are responsible for those choices. That includes complying with the laws, professional rules and organizational policies that apply to your work. Organizations deploying Ethen should set their own rules for when and how AI output may be used.

If you see Ethen produce something harmful, report it through the contact options in the Legal & Trust Center.