ONLYOFFICE 10.0 AI: Open-Source Builds Skip It, XSS Bug Patched

What shipped

ONLYOFFICE has released Docs 10.0, announced on October 10, 2026. Techaeris calls it the suite's biggest autumn release in years, with around 50 new features and more than 1,000 fixes across the online, desktop, and mobile editors 3. The headline feature is AI built directly into the editing experience. A dedicated AI tab with an integrated chat lets users draft, summarize, translate, clean up, and format content without leaving the document or pasting text into another tool 3. Because the assistant is context-aware and reads the file it is working in, its output should be more relevant than a general-purpose chatbot's 3.

The release also changes how the suite is organized. Docs now opens to a home screen with a "Discover Apps" section listing four free, pre-installed modular apps: Files, Rooms, Forms, and AI agents 1. Each one is optional and meant to be used only when needed. Files, for example, handles storing, organizing, and sharing documents 1.

The open-source caveat

The most notable detail is easy to miss. ONLYOFFICE is adding AI tools to every editor, but the open-source versions do not include them out of the box 1. According to It's FOSS, the company says it made this choice in response to user feedback, so those builds require AI tooling to be installed separately 1.

That framing matters. Techaeris describes ONLYOFFICE as an open-source office suite and presents the AI as its centerpiece 3. It's FOSS, writing for a Linux and FOSS audience, puts the exclusion near the top of its coverage 1. Both accounts can be accurate. The AI is a headline feature, but whether you get it by default depends on which edition you run.

There is a reasonable case for the approach. Many people who choose open-source office software do so because they want control over their data, and some actively distrust AI features that send document content to outside services. Making AI opt-in for the community builds respects that preference. The cost is that "AI in every editor" is a less complete promise than the marketing suggests, and administrators of self-hosted, open-source deployments should expect to do extra integration work to match what commercial users get by default.

AI brings new attack surface

The security side of the release shows the trade-off clearly. ONLYOFFICE's own blog says version 10.0 contains more than 40 security fixes across the online editors and the desktop app. They cover document conversion, PDF handling, macros, integrations, and server configuration 2. One fix addresses a cross-site scripting (XSS) flaw in the Value field of the new AI() function, which could be exploited to make the editor stop working 2. XSS involves injecting a malicious script into content that other users then load 2.

The reported numbers differ. Techaeris cites 35 security fixes 3, while the vendor's community post says more than 40 2. The vendor's figure is more likely to be current, but either way the release is substantial hardening work and not a cosmetic update.

The AI() flaw is the most instructive item on the list. When a function moves model-driven or user-supplied text into a document environment, it becomes a path for injection. The impact ONLYOFFICE disclosed is denial of service, meaning the editor stops working, not data theft 2. That is relatively contained. Still, it shows that embedding AI in productivity software is not only a feature decision. Every place where AI input or output is rendered has to be treated as untrusted.

Why it matters

ONLYOFFICE is joining the broader move to put AI writing and summarization tools inside the documents where work happens, instead of in separate chat windows 3. For teams evaluating alternatives to the large commercial suites, the pitch is now stronger: modular apps, context-aware assistance, and security fixes aimed at enterprise buyers 3.

The more interesting story is how ONLYOFFICE handled two pressures at once. It made AI central to its commercial offering while keeping it out of the default open-source builds, citing what users asked for 1. It also shipped a fix for a vulnerability that exists only because the AI functionality was added 2. On this reading, ONLYOFFICE treats AI as a feature to sell and also as a risk to contain. That is a more careful position than many competitors take, even if it leaves open-source users with a more fragmented experience.

The practical advice follows from that. Commercial customers should update promptly to pick up the security fixes. Open-source administrators should decide deliberately whether to add AI at all, and if they do, they should treat that integration with the same caution ONLYOFFICE's own patch notes imply.