The Document Foundation writes down six conditions for adding AI to LibreOffice, and says none are met yet

The Document Foundation has turned LibreOffice's refusal to ship AI features into a written product requirement, publishing six conditions that any AI integration must clear before it ships by default — and, as the foundation stated again this month, none currently does. The policy first appeared when the nonprofit released LibreOffice 26.8 in late August, under the header "no artificial intelligence." It resurfaced this week when TechCrunch covered it as a rare case of a software maker marketing the absence of AI as deliberately as competitors market its presence.

That framing is the part worth sitting with. LibreOffice isn't claiming it skipped AI because building it is hard, or because the roadmap hasn't gotten there yet. It's claiming the absence is a choice, governed by criteria it's willing to publish and be held to. For an industry where nearly every roadmap slide now has an AI line item, a vendor arguing in public for why it hasn't added one is the unusual move.

A policy, not a slogan

The trigger was a question. After the 26.8 release, a community member asked the obvious one: "So, no AI is now a feature?" Rather than answer with a line for the press, Italo Vignoli, the foundation's spokesperson, used a September 3 blog post to lay out exactly what would have to be true before LibreOffice shipped an AI feature by default. The list reads less like a mission statement than a procurement checklist:

• Users choose where inference runs — their own machine, infrastructure they control, or a service they pick themselves
• No document content leaves the computer without the user's explicit authorization
• No telemetry of any kind, matching the suite's existing no-data-collection policy
• No integration that works with only one company's API
• Generated content stays in ODF format, with structure and styles intact
• Any AI feature remains fully optional, installable and removable

None of the AI integrations available today, the foundation said, meet all six at once. Until one does, LibreOffice ships with no AI of any kind in its default installation. Users who want it can still add a third-party extension that connects to a locally run model, which keeps the document from leaving the machine at all.

Why a nonprofit can say this and a vendor usually can't

The more useful part of the post for marketers isn't the privacy argument — it's the foundation's account of why competitors reach a different conclusion.

"An integration that only works with a single company's API is a lock-in mechanism, regardless of how it is described in the release notes."

That line is aimed squarely at the standard SaaS pattern: an AI assistant bundled with a single model provider, sold as a feature, that also happens to justify a price increase and keep documents inside one company's infrastructure.

The Document Foundation can afford to wait because it has no subscription tier to protect and no upsell riding on an AI line item. Most B2B software vendors don't have that luxury — AI features are already priced, and in many product lines, mandatory. That's exactly why LibreOffice's position is being covered as news rather than ignored as a nonprofit's technical quirk: it's a visible counterexample at a moment when "AI inside" has become close to universal in software marketing, and some buyers are starting to ask what that inclusion actually costs them in data control.

The foundation names its own highest-stakes users directly: schools, hospitals, public bodies and law firms, the kind of customers whose documents routinely qualify as legally protected material. Those buyers don't experience an AI assistant as a convenience add-on. They experience it as a new place their compliance team has to sign off on, a new vendor relationship to audit, and a new line in a data processing agreement. A feature that can't answer basic questions about where it runs is a liability in that buying process regardless of how well it performs.

What this means for your positioning

Those are the same buyers software marketers increasingly have to answer to, regardless of whether the product in question is a document editor. If your AI feature can't answer where inference happens, whether content leaves the customer's environment, and whether the integration can be removed, a security-conscious prospect will eventually ask anyway — and "it's AI, so trust us" is not an answer procurement teams accept anymore. LibreOffice's checklist is, in effect, a preview of the questions now showing up in security reviews and vendor questionnaires across regulated buyers:

• Where does inference actually run, and can the customer choose or restrict it
• Does any document content leave the customer's environment, and under what authorization
• Does the feature depend on exactly one model provider, creating a new lock-in point
• Can the feature be fully disabled or removed by an administrator, not just hidden from the interface