Build · founder · 7 min read

Lovable Starts Training on Your Projects September 9. Here's the Toggle.

On Sep 9, Lovable may use Free and Pro customer data — prompts, code, project files — for model training unless you opt out. Business and Enterprise are exempt.

Published September 2, 2026 ·

lovable privacy model training data founders

Seven days from now, on September 9, 2026, Lovable’s updated privacy policy takes effect. From that date, customer data on the Free and Pro plans may be used to train, develop and improve Lovable’s AI models — unless you have turned a toggle off.

The change was published in Lovable’s changelog on July 31 and sits in the docs under a page most people will never open. It is not hidden, exactly. It is just not the kind of thing anyone reads on a Tuesday.

If you have built anything on Lovable’s Free or Pro tier, this is the one thing worth doing this week.

What “customer data” actually covers

Lovable’s definition is broad, and it is worth reading slowly. Customer data means the content you submit to or create with Lovable:

• Prompts — including any images and files you attach
• Code generated in your projects
• Project files and configurations
• Hosted applications
• Generated outputs
• Usage data

So: the description of your business model you typed into chat to get the schema right. The competitor screenshot you pasted in for a layout. The pricing logic. The half-finished onboarding copy. The whole thing.

Two categories are explicitly excluded, and the distinction matters more than it first appears.

Your app’s end-user data is not used for training. Whatever your customers submit stays in your project’s own database and storage. If you run a booking app, your bookers’ details are not in scope. That is the right line, and Lovable drew it clearly.

Your account and billing details are not used for training either. They run your account, nothing more.

The plan split is the real story

Here is the part that should give a bootstrapped founder pause.

Business and Enterprise workspace data is excluded from model training by default. No toggle, no action, no deadline. It is governed by the organisation’s Data Processing Agreement and simply sits outside the training set.

Free and Pro users are opted in by default and must switch it off themselves.

That is a defensible commercial arrangement — enterprise contracts carry data terms that individual subscriptions don’t, and Lovable is honouring them. But look at what it produces in practice. The people whose data trains the model are the ones building the earliest, scrappiest, most original things: the solo founder testing an idea at 11pm on the Free tier, the two-person team on Pro building the first version of a product nobody has built before.

The companies that can afford to be excluded are excluded. The companies that can’t, aren’t. Worth being clear-eyed about, whatever you decide.

How to opt out

It takes about thirty seconds and it costs nothing.

1. Open Account settings → AI model training in Lovable
2. Disable Use my Lovable content for model training

Lovable states plainly that opting out is free and does not affect your use of AI features. There is no degraded tier, no nag screen, no feature you lose. That deserves credit — plenty of vendors make the privacy-preserving option quietly worse.

Two details about how the setting behaves:

It’s account-level, not workspace-level. The setting covers the content you contribute. It does not change the setting for anyone else in a shared project or workspace. If you have a co-founder or a contractor in the same workspace, each of you controls your own, and you need to tell them. This is the single most likely way a team gets this wrong.

If every workspace you belong to is Business or Enterprise, the toggle appears already off and locked. Nothing to do.

Why the date matters, and not just the toggle

You can opt out before or after September 9. But the two are not equivalent, and Lovable says so directly.

Opt out before September 9 and training under the new policy will not have started, so your content is not used for training at all.

Opt out after and your content is excluded going forward only. It does not retract anything from training datasets already assembled or models already trained.

That is a one-way door with a date on it. Not a catastrophic one — but there is no version of “I’ll get to it next month” that gets you the same outcome as doing it this week.

The second change nobody is talking about

Buried in the same policy update is a separate item that has nothing to do with model training.

From September 9, Lovable may share limited pseudonymized identifiers with advertising platforms. Lovable is explicit that it does not share the contents of your projects — prompts, code, or your app’s end-user data — with advertisers, unless you instruct it to. The identifiers are a different thing: signals that let ad platforms recognise you across services.

The regional handling is what you’d expect from a company that has done its homework. In the EEA, UK, Switzerland and Brazil this happens only with your consent. In the United States you can opt out at any time via the “Do Not Sell or Share My Personal Information” link in the footer, your privacy settings, or a Global Privacy Control signal.

If you are a US-based founder, that is a second, separate opt-out, in a different place, with a different mechanism. Do both while you’re in there.

What this fits into

This is the fifth time this year a major AI coding vendor has changed something material without changing a price.

We’ve tracked the allowance shrinking, the metering unit changing, the routing default moving, and last week, retention windows extending. Add training rights to the list.

The pattern is consistent enough now to plan around. The sticker price is the most stable number in your AI stack and the least informative one. What moves is everything the price doesn’t cover: how much you get, how it’s counted, what’s on by default, how long it’s kept, and now what it’s used for afterwards.

The practical version: once a quarter, open the account settings of every AI tool you pay for and read the toggles. Not the pricing page — the toggles. That is where the terms of your relationship with the vendor actually live, and it is the only place you get a vote.

Do Lovable’s this week.

---

Sources read directly: Lovable — Manage training data and privacy, Lovable changelog, July 31 2026 entry, Lovable Privacy Policy (version 2026-09-09).