Figma AI training opt out: who is on by default, and the switch Framer added in September

If your team designs in Figma on a Starter or Professional plan, the Figma AI training opt out is a switch someone has to go and find: Figma's help center says content training is on by default on both tiers, while Organization and Enterprise are off by default. Framer, Sketch and Adobe each answer the same question differently, and two of those answers changed this year.

This post sets out what each vendor's current document says, with the date on it, so you can tell in about ten minutes whether your files are feeding a model and what you can actually do about it. We checked Figma's, Framer's, Sketch's and Adobe's own pages on October 23, 2026. Where a vendor's page would not load for us, we say so rather than guess.

Figma content training is on by default for Starter and Professional teams

Figma's help center article on AI features and content training settings is explicit about the split:

• Starter: content training is turned on by default. Team admins can turn it off at any time.
• Professional: on by default, and again only team admins can change it.
• Organization and Enterprise: off by default. Figma's page says content training "is currently disabled for Organization and Enterprise plans and can't be toggled on right now."

Two details in that article matter more than the toggle itself. First, drafts that sit inside a team follow that team's setting, so moving a draft into an opted-in team opts the draft in. Second, Figma for Education and Figma for Government data is not used to train AI models at all.

The article also separates two settings people conflate. Content training decides whether your work trains models. AI features decides whether your team can use Figma's AI at all. Turning content training off does not remove the features, and paid AI credit limits are managed separately, per person.

Figma's own account of what it trains on is narrow: models that learn design concepts and patterns, plus Figma's internal file formats and structure. The help center says Figma de-identifies content and redacts sensitive information so that models learn general patterns rather than a customer's specific concepts. Content training took effect on August 15, 2024.

Figma is one of 374 products in Upfynd's design tools category as of October 2026, and in our view it is the one to check first, because the default runs the other way on its two self-serve tiers.

How to use the Figma AI training opt out, step by step

The path differs by plan, and in both cases you need admin rights. From Figma's help center, as of October 2026:

1. On Starter: open your team in the file browser, click All Folders, select the icon next to the team name and choose View settings. Go to Settings, scroll to the AI section, select Manage AI settings, then toggle Content training.
2. On Professional: open your team in the file browser, click Admin, go to Settings, scroll to the AI section and toggle Content training.
3. On Organization or Enterprise: there is nothing to do. The setting is off and cannot currently be switched on.
4. For drafts outside a team: Figma's FAQ answer is to turn content training off, or move those drafts into a team where it is already off.

If you are not an admin, you cannot do this yourself, and neither can an individual account holder opt out of a team's choice. In our view that is the single most important line in the whole subject: on Professional, your personal preference is not a setting that exists.

What switching it off does not do

Figma's AI Terms, effective June 24, 2026, scope the opt-out tightly. Section 4 says that "new Customer Content and edits added after the date the toggle was turned off will not be used to train AI models." It is forward-looking. Neither the AI Terms nor the help center article we read describes removing or unlearning content already used, and we would not assume either happens.

The same terms are clear on ownership: "As between Customer and Figma, Customer retains all right, title, and interest in Customer's Input and Output." They also disclaim accuracy - Figma "makes no warranty or guarantee as to the accuracy, completeness, or reliability of Output" - and put the job of evaluating output on you. There is no output indemnity in the AI Terms themselves.

One more limit worth stating plainly: the toggle governs your files, not the licences on everything you pull into them. Stock photography, icon sets and templates come with their own terms from whoever published them, and a design-tool opt-out does not reach them. If that matters for your brand work, the licence to read is the one attached to each asset library you use from the design resources category, not your editor's privacy page.

The question is contested, not settled. The OECD's AI Incidents Monitor records a California class action dated November 2025 alleging that Figma used proprietary user-uploaded data without consent to train generative AI models, and that it contradicted promises not to use customer data for training. Those are allegations in a filing we have not seen adjudicated; we are reporting that the case exists, not that the claims are true.

Framer quietly added a workspace opt-out in September, and the old wording is still being quoted

This is the change we saw third-party guides miss while researching this post. Framer's AI Notice is versioned, and the two most recent versions say different things.

Version 1.1, effective June 16, 2026, granted Framer a licence with no opt-out for anyone below Enterprise: "All non-Enterprise Plan customers hereby grant Framer a worldwide, limited, irrevocable, non-exclusive right and license to use and reproduce Customer's Inputs and Outputs in a deidentified manner for the purposes of training Framer artificial intelligence models."

Version 1.2, effective September 18, 2026, opens that sentence with a condition: "Unless AI model training is disabled in the applicable Workspace settings, Customers on a non-Enterprise plan grant Framer a worldwide, non-exclusive license to use their Inputs and Outputs to train." Enterprise is excluded in both versions - section 7.2 says Framer "will not use the Inputs or Outputs of Customers on an Enterprise plan to train Framer's AI models" - and section 7.1 says none of Framer's third-party model providers has the right to use customer inputs or outputs to train.

So a non-Enterprise Framer workspace now appears to have a switch that it did not have three months ago. Two honest caveats: the notice does not say where in Workspace settings that control sits, and the superseded version 1.1 is still what several privacy-tool blog posts quote when they tell you Framer gives you no choice. If you read somewhere that Framer's licence is irrevocable, check which version the writer was looking at.

Sketch does not claim to train on your documents, but its MCP server moves the risk

Sketch's privacy statement, labelled version 06 October 2026, does not say Sketch uses customer documents to train its own models. It mentions training only in relation to other companies' tools: if you connect a third-party AI tool, "that provider's privacy policy governs how it processes, retains, and may use this data, including for model training."

Two specifics from that privacy statement are useful. Background removal "processes images locally on your Mac using Apple's on-device machine-learning frameworks," so those images are not sent anywhere. And Sketch's Model Context Protocol server - the bridge that lets an AI client read and edit design files - is off by default. Turn it on and the exposure is real, but it belongs to whichever AI vendor you connected, not to Sketch.

That pattern now applies well beyond Sketch. MCP is a shared protocol, so the same reasoning holds anywhere you wire a design tool to an assistant: the relevant policy is the assistant's. A clean vendor privacy page says nothing about the model on the other end of a connection you opened yourself.

Adobe's policy answers a narrower question than people think

Adobe's privacy policy, last updated October 24, 2025, says Adobe "may perform analytics with Cloud Content to help us understand how our users are using our Services and Software," and makes that "subject to your opt-out rights." What the policy we read does not do is state explicitly whether generative models are trained on Cloud Content.

Adobe's separate generative AI user guidelines say prompts, inputs and generated results "may be reviewed through both automated (such as machine learning) and manual methods" "for abuse prevention and content filtering purposes" - review, which is not the same thing as training. If you need a contractual answer for Adobe rather than an inference, the place to get it is your own agreement or a written answer from Adobe, and we would ask for it in writing before promising a client anything.

Who actually has to act

Two gaps we will not paper over. Adobe is missing from that table because the policy we read does not give a default for generative training, only for analytics. And Canva is missing because Canva's trust and help pages returned HTTP 403 to our requests on October 23, 2026, so we could not read the current wording; its privacy settings page inside the product is where to check, and we would rather send you there than repeat a number from a third-party summary. Our earlier look at what Canva's first image model is good for covers the features, not the training terms.

If you manage a design team, the five-minute version is this: open the admin settings of every design tool your team pays for, screenshot the training toggle with today's date, and put a calendar reminder to look again in six months. Defaults change quietly: Framer's two versions differ, and we found no announcement of the difference. Then compare what else is out there in the design tools category or the wider privacy category if a vendor's answer is not one you can live with.