LibreOffice 26.8 is here… guaranteed to be free of generative AI.
The Document Foundation had set the tone at the end of August with the launch of this new version. It argued for a “deliberate position,” essentially aimed at preserving privacy. It also recalled, on this point, that no component of the suite required network access.
Only a few days later, the foundation offered further details. It said it was not categorically opposed to AI integration, but explained that it would refuse to include it as long as six conditions were not met. In broad strokes:
- The user must be able to decide where inference takes place
- No content may leave the local device without permission
- No telemetry, as is the case for the entire LibreOffice suite
- No reliance on a single provider
- Generated content must conform to the ODF format
- Any AI features must be removable and their UI hidden for those who do not wish to use them
Under these terms, no integration could be enabled by default in its current state, The Document Foundation concluded. It did remind users that they can install extensions themselves—most of which rely on local models.
Read also: Collabora and LibreOffice at odds: what is at stake after years of conflict
OnlyOffice no longer preloads AI on its open-source versions
At OnlyOffice, the approach is the opposite for AI capabilities. Specifically on the cloud and server editions. User feedback has led the company to reverse the approach on open-source versions: AI is no longer preloaded, and the tools are now installed separately.
iA Writer ditches automated drafting and highlights text generated by the LLM
To date, Information Architects had avoided embedding automated writing features in its iA Writer app. It has, however, integrated a mechanism that distinguishes content generated by large language models by graying it out.
Visible only to the author, the mechanism relies on a Markdown convention that preserves the distinction when text is copied into another software. Launched late 2023 on Mac, iOS, and iPadOS, it arrived this year on Windows.
A “delegation of judgment”: Vivaldi rejects any AI assistant
In the summer of 2025, Vivaldi Technologies took a firm stance again: “Keep browsing human.” The company explained it had no intention of embedding a large language model to add a chatbot, a summarization tool, or a recognition engine to its browser—at least as long as there was no way to provide these services “in a robust manner.” “If an AI proves useful without infringing on intellectual property or compromising privacy or the open web, we will use it,” the statement concluded. “Not if it makes us passive consumers.”
Early 2026, the message had not changed. Vivaldi Technologies reiterated the absence of an AI assistant in its browser. “When an artificial assistant interposes itself between you and the web, it filters what you see and decides what you do not see. That is an externalization of judgment.”
Procreate says no to GenAI for ethical reasons
Savage Interactive develops Procreate, the digital art app for iPad. It had backed away from its stance on generative AI in the summer of 2024. The takeaway: there will be no GenAI in its products, for ethical reasons and to safeguard human creativity (which “is shaped by learning and gesture”). This does not prevent the inclusion of machine-learning algorithms for features such as shape recognition and stroke stabilization.
Clip Studio Paint discouraged by its community
Savage Interactive also states that it does not use GenAI internally.
Its competitor Celsys had planned to integrate an image generator into Clip Studio Paint in November 2022. It quickly reconsidered after community backlash, largely for the same reasons as Procreate: potential IP violations and restrictions on creativity.
In a few months, Sketch eased its position
In September 2024, Sketch, the Mac app of the same name, was not yet ready to take the leap. “AI may be our future, but it will not generate designs,” it had stated, viewing it instead as an aid for tasks like renaming layers, organizing files, and isolating images. Its principal ethical argument was that it would only rely on third-party models under conditions of transparency about training data and permission to use them.
By 2025, the situation had shifted significantly. Sketch had equipped its application with a local MCP server. It remains exclusively local and is not enabled by default.
No native AI at 37signals, but open to agents
37signals, known for taking firm stances on certain technology issues, has integrated a section on AI into its privacy policy. It asserts that it has not integrated features that process user content in its products.
These assurances appear in response to the HECVAT (Higher Education Community Vendor Assessment Toolkit), a self-assessment questionnaire for security and compliance aimed at technology vendors selling services to secondary and higher education institutions. When asked, “Does your solution have AI features or do you plan to implement them in the next 12 months?” 37signals answered no for Basecamp, its flagship product. It also notes, however, that it allows connecting agents via its CLI and MCP servers.
