Responsible Use
How to use AI ethically
Not principles. Practices: specific things to do, each one traced to reporting on the record.
Every practice here links to the wire items behind it. When the record moves, this page moves with it. A practice
we can't source doesn't appear, however sensible it sounds.
Last checked against the record 2026-10-17.
If you use AI
Turn off training on your chatbot conversations
Most major chatbots train on your conversations by default. In ChatGPT, turn off "Improve the model for everyone" (Settings, Data Controls); in Claude, "Help improve Claude" (Settings, Privacy); in Gemini, Gemini Apps Activity; in Meta's Muse, "Help improve our AI" (Settings, Data controls). The ChatGPT switch covers only future chats.
Why No U.S. federal law guarantees a right to opt out, so the settings page is the protection you have.
On the record: How to opt out of AI training in ChatGPT, Gemini, Claude and Meta Muse
Keep your smart glasses' camera out of AI training
In the Meta AI app, open each pair's glasses settings, choose "Meta AI" and turn off "store visual data from AI experiences," with the glasses out of their case and connected. Voice isn't covered: delete recordings yourself.
Why Otherwise what the camera sees can train Meta's models and be reviewed by contractors.
On the record: Meta adds a setting to keep smart-glasses camera requests out of AI training
If your organization adopts AI
Never treat an AI detector or watermark check as proof
Don't cancel a contract, fail a student or discipline anyone on a detector result alone. A missing watermark doesn't prove a person wrote it, and light editing defeats detection. If you use detectors, say so in a written policy, and get permission before uploading someone's work to one.
Why OpenAI's own tests show its watermark fading after synonym swaps, and the Authors Guild says publishers have already canceled books over alleged AI use.
On the record: Authors Guild urges publishers not to cancel books on AI detector results alone · OpenAI to watermark ChatGPT and Codex text in the EU under the AI Act
Judge authorship by the record, and give people a hearing
Ask for work in a format that keeps full version history, and before acting on suspected AI use, ask the writer about it and offer an appeal.
Why A record of how something was written is fairer evidence than a hunch or a score.
On the record: Case Western student paper The Observer requires version history to police AI use
Assume anything you upload to an AI tool can be kept and trained on
Never put confidential, source or client information into an outside AI service. When generative AI plays a significant part in published work, disclose the tool and how much it was used, and keep a human check on every AI-assisted step.
Why A vendor's privacy promise can change; your own rule doesn't have to.
On the record: RANGE Media's newsroom AI policy bans AI-written content, requires disclosure
Adopt a written AI policy, starting from a vetted template
If your school, newsroom or organization has no AI policy, adopt one rather than leaving rules to each classroom or desk. You don't have to start from scratch: many New York districts adapted their state school boards association's sample language.
Why No policy means everyone improvises, and the people with least power absorb the mistakes.
On the record: Most New York school districts have no AI policy, New York Focus finds · RANGE Media's newsroom AI policy bans AI-written content, requires disclosure
Keep a licensed human as the decision-maker when AI informs care
Clinicians should keep independent judgment when decision-support tools inform patient care, and no AI tool should independently perform a clinical function reserved for a licensed professional. Identify, mitigate and monitor tools with a risk of biased impacts.
Why California now requires this; any health system can adopt it without waiting for a law.
On the record: Newsom signs California law requiring clinicians' own judgment when AI informs care
Put AI governance in front of the board
Treat AI as a board-level matter, with written expectations for governance, risk management, and trust and transparency, as Hong Kong's bank regulator now asks of banks.
Why When AI starts acting on its own, accountability has to sit with someone who can stop it.
On the record: Hong Kong Monetary Authority issues ART responsible AI guidance for banks
If you build AI
Get consent for every use of a person's voice, and pay for it
Ask before training on, generating or publishing someone's voice, say what it's for, and pay according to how it's used. Don't penalize performers who refuse or withdraw consent.
Why Japan's performers wrote these rules themselves because the law there recognizes no right to one's own voice. Ethics shouldn't wait for the law.
On the record: Japan Actors Union guidelines require consent at every step of AI voice use
Label AI-made content, and say how much AI made it
If you host or publish media, label it "fully," "partly" or "not" made with generative AI, and show that label where people actually listen or watch.
Why People can't weigh what they're hearing if they don't know a model made it.
On the record: YouTube Music starts labeling songs made with generative AI