ChabadLabs AI FOR SHLICHUS
EN

Chassidus Copyright and Access

The 2026 debates over who may digitize, OCR and open up the Rebbeim's seforim for AI, what each side argued, and what a builder can use today.

On this page

An AI tool can only quote a sicha accurately if it can read the sicha. Most of the Rebbeim's seforim are not available to general AI tools as clean text, and the publishers hold copyright in the editions, translations and typesetting. In 2026 this became the hardest argument in the builders' group: should the texts be opened up so anyone can build on them, or kept under the publishers' care so their accuracy is protected?

This page records what each side said. It does not rule on the halachic questions involved (copyright in halacha, hasagas gvul, the publishers' rights). Ask your rav about those.

Why it matters for AI

The Torah accuracy problem and the access problem are the same problem seen from two sides. A model that has never seen the text will invent a citation. A builder who cannot get the text cannot build a tool that cites it correctly. In March 2026 a builder noted that a year earlier there had been broad agreement that a Chassidus search-and-answer platform was the next step, and none existed publicly, partly because builders were waiting for the publishers.

The March 2026 debate

On March 22 and 23, 2026, the builders' group argued for two days. The group admin opened with "We need Sefaria of Chassidus," and later told the main group the conversation was "a key to unblocking the next frontier in hafotzas hamaayanos."

The case for opening the texts

  • Printing is one thing; digital text is another. Copyright can stay in place while the text is open for apps to use.
  • Sefaria and Chabad.org already host copyrighted texts with permission and a notice.
  • If the printer's files were released, many volunteers could fix OCR errors quickly.
  • Builders are stuck: a shliach's Purim site of AI summaries of real sichos stayed private because the text was not licensed.
  • Proposed terms: a non-commercial Creative Commons license, the publisher's logo and attribution in every app, and a link to buy the sefer.

The case for careful stewardship

Relayed by a member connected to SIE:

  • The worry is that the Rebbe's words and hashkafa "become hefker with no oversight," and that AI can be "great and be delusional" at once.
  • People need "a place which we feel confident in its authenticity."
  • A licensing option to use the content as a base was planned, and all the seforim would eventually be online.
  • No publisher lets its name be used on a product it does not control.

The middle position

  • License the text with oversight and the ability to revoke access, the way some free APIs still require an application.
  • Let one trusted body build the platform and open an API, and let shluchim adapt it locally.

The OCR side argument

Members disagreed about whether AI can clean up scanned Chassidus:

  • One view: much of Kehot's library was never digitized, old typesets are hard to scan, and "you absolutely cannot put out a Tanya with a typo." A member reported SIE was building a page-by-page review where a human approves small fixes in seconds.
  • Another view: a good model corrects most OCR mistakes because they "don't make sense."
  • The reply: correcting "based on meaning" is guessing, and "I would not rely on that at all."

That night one builder made a proof-of-concept crowd-checking site: each scanned line goes to three volunteers and is accepted only if all three agree. Other ideas included showing the scan next to doubtful OCR text, or laying OCR text over the scan in the same font and toggling between them to spot errors. A separate 2026 tip for Hebrew pages: split the page and read the right half before the left, because AI often garbles column order.

Members on both sides also drew on the Rebbe's approach. Some stressed urgency in spreading the wellsprings; others pointed to the Rebbe's protectiveness of Kehot's seforim and the copyright notice inside Likkutei Sichos. The stories told on each side were not sourced in the chat.

The September 2026 debate: Daiah's copy block

Daiah, the publishers' AI and text platform, entered closed beta in July 2026. Its reader does not allow copying. On September 22, 2026 the group admin criticized the block sharply. People involved with the platform answered:

  • The platform is not the publishers, and the publishers decide reuse terms.
  • "The fact that certain seforim are online in plain text at all is huge."
  • More uses were coming.

A shliach added a concrete case: he needed Chabad siddur text for Hebrew school davening and could not copy it anywhere until a Chabad siddur reached Sefaria.

By September 23 a builder who had been critical said he had "come around" and was pleased with Daiah's progress. On September 18 another member observed that Chassidus builders feel a copyright "chilling effect" that keeps them from sharing what they are working on.

On September 25 members noticed the book text had disappeared from chabadlibrary.org, a volunteer site whose text comes from dach.dev. The same day, a member relayed that the library's director called it a glitch being fixed. The lesson the group took: do not build on a single volunteer-run text source without a backup plan.

What the rules say now

Chabad.AI's Terms of Use (effective October 13, 2026) are the clearest statement from the publishers' side:

  • Allowed: reading, searching and studying, and sharing "individual answers and short excerpts with their source citations for personal, educational, and non-commercial purposes."
  • Not allowed: copying, downloading, scraping or extracting the library's texts; republishing them; or using them "to train, fine-tune, or build an AI model or a competing service."
  • For anything beyond that, the Terms invite you to contact Chabad.AI through the support address they list.

What a builder can use today

Text that is openly licensed or offered for reuse, as of October 2026:

  • Sefaria. Kehot's Tanya (Hebrew and English), Torah Ohr, Likkutei Torah and Derech Mitzvosecha arrived in December 2022 under a non-commercial license (CC-BY-NC). Sefaria also carries a Chabad weekday siddur, with Hebrew text drawn from Wikisource. Sefaria's Texts MCP lets Claude or ChatGPT search all of these with citations.
  • dach.dev. A plain-text Chassidic library many builders use for personal study tools. Check its terms before you publish anything built on it.
  • Chabad.org. Many texts and translations in its Torah Texts section, for reading. Permission is a separate step: in February 2026 the megillah.app builder asked and received the go-ahead to use the Kehot translation of the Megillah, and members said Chabad.org's editors usually say yes to community projects that ask.
  • Daiah. For reading and asking questions, not for extracting text.

If a project needs more than that, ask the publisher before you build. A short email that explains the project, the audience, and how you will credit and link back gets further than a finished app that has to come down.

Last updated October 2026 · Maintained by ChabadLabs AI Agent