Anthropic has added a gallery of ready-made artifacts to Claude for Legal, announced on October 28, 2026. Anthropic's own definition: "Artifacts are interactive tools, such as trackers and dashboards, that teams can use and share in Claude." The legal set includes a cite checker, a jurisdiction survey matrix, a matter board, a playbook deviation board, a regulatory horizon scan and a tabular contract review, with outputs that link back to their sources. The Decoder frames the headline as verification: the model checks sources first, then hands the file to a lawyer.
This is the practical follow-up to the product Anthropic launched in May. Below: what is in the gallery, what the verification step does and does not do, and how to test it before trusting it with a filing. For wider context on why citation checking matters, see our explainer on why lawyers keep getting sanctioned for AI hallucinations.
TL;DR: what is in the release
| Question | Answer (per Anthropic's announcement) |
|---|---|
| What is new? | A gallery of task-specific artifacts inside Claude for Legal |
| Which artifacts? | Cite checker, jurisdiction survey matrix, matter board, playbook deviation board, regulatory horizon scan, tabular contract review |
| Connectors | 60+ legal connectors added since May, including Bloomberg Law, Filevine, Ivo, Legora |
| Partner artifacts | Goodwin, Harvey, Foley & Lardner, PwC |
| Access to justice | Nine artifacts, five built by nonprofits that received Claude credits |
| Community | Funding for public interest groups and legal education; a Legal Ambassadors Program with a waitlist |
| Pricing | Not published; contact sales or an implementation partner |
What are artifacts, and why legal teams asked for them
Artifacts in Claude are interactive outputs, like a small app or dashboard, that live beside the conversation. We covered the general feature, and how connectors feed it, in Claude artifacts and MCP connectors. The legal gallery packages that idea into templates for jobs lawyers repeat daily.
The Decoder reports that legal teams had requested these since the May launch. That makes sense: a chat transcript is a poor object for a matter team to share. A board that shows each matter's status, or a matrix comparing statutes across jurisdictions, is something a partner can open, filter and send to a client.
The cite checker and the "verify first, lawyer second" workflow
The most discussed piece is citation verification. Per The Decoder's summary, the model performs the initial verification and links each statement to its original source; a lawyer then reviews the results. Anthropic's gallery description says outputs link back to their sources.
Why that structure matters: the failure mode that has produced court sanctions is a citation that looks right and points to nothing. A tool that attaches a source link to every statement lets a reviewer click and read, instead of re-searching from scratch. It does not make the claim true. If the linked source does not say what the draft says, only a human reading it will notice.
A sensible protocol for any team using the feature:
- Treat every machine-checked citation as unverified until a person opens the source.
- Spot-check at least a sample against an authoritative database, not just the link the tool provides.
- Keep a log of what was checked, by whom and when, in case a court or client asks.
- Never let the tool's pass/fail label stand in for the lawyer's sign-off.
Our fact-check of a related case, a paper that cited a TorrentFreak article that does not exist, shows how convincing a fake source can look.
Connectors and partner tools
Anthropic says it added more than 60 legal connectors since May, naming Bloomberg Law, Filevine, Ivo and Legora. Users can connect their own databases without manual uploads, according to the Decoder. Goodwin, Harvey, Foley & Lardner and PwC contributed artifacts of their own.
The presence of Harvey is notable. Harvey sells a competing legal AI product, and appearing as a contributor suggests the legal market is treating Claude as a platform to build on as well as a rival. OpenAI has its own pitch for lawyers, which we covered in OpenAI's Astra for law.
Data controls: what Anthropic claims
The Decoder reports that Anthropic says built-in settings enforce data protection rules and maintain the information barriers clients require between teams. That is a significant concern in law firms, where ethical walls separate matters. We could not find technical documentation of how those barriers are enforced, so a firm should ask for it in writing and test with a staged pair of matters before relying on it.
Access to justice and community programs
Nine artifacts in the gallery target access to justice, and five of them were built by nonprofits that received Claude credits. Anthropic is also funding public interest groups and legal education and starting a Legal Ambassadors Program with a waitlist. These are modest but real signals that the company wants legal-aid users, not only large firms.
How the six artifacts map to real legal work
It helps to read the gallery as a list of recurring jobs rather than a feature list.
- Cite checker: the pre-filing chore. A junior associate traditionally pulls every authority in a brief and confirms it exists, says what the brief says and is still good law. The artifact aims to do the first part and attach links for the rest.
- Jurisdiction survey matrix: a table comparing how several jurisdictions treat the same question. Anthropic's pitch is comparing laws across countries, which is slow to build by hand and easy to get subtly wrong when a jurisdiction has amended a statute.
- Matter board: case file management as a shared dashboard, a Kanban-style view of deadlines and status per matter.
- Playbook deviation board: checks contracts against a team's internal guidelines and shows where a counterparty's draft departs from your standard positions.
- Regulatory horizon scan: tracks proposed legislation, so a compliance team sees what is coming and not only what has passed.
- Tabular contract review: extracts clauses from a stack of agreements into a comparable table.
Each is something firms already pay people or other software to do. The competitive question is not whether Claude can produce such a table once, but whether it is reliable and auditable enough to run daily, which is what the connector count and source links are trying to address.
Where this fits in the legal AI market
Legal is one of the first professions where AI vendors sell directly against specialist startups. Anthropic launched Claude for Legal in May; OpenAI has pitched its own models for legal research, and companies like Harvey build on top of several models. The fact that Harvey and large firms contributed artifacts to Anthropic's gallery suggests many of them prefer to ride a general platform than to compete with it on every layer.
For buyers that has a practical consequence: pricing power is shifting from specialist wrappers toward the model providers, and firms should expect more bundling. For vendors, the opportunity is distribution: an artifact in a gallery reaches every Claude Enterprise legal team at once.
The risk side is unchanged. The legal profession has the clearest public record of AI failure, with courts publicly sanctioning filings that contained invented cases. Any product in this space will be judged less on demos than on the first widely reported miss. That is why the human review step in Anthropic's own framing is not a footnote but the product's safety design.
What to test before you deploy
- Citation accuracy: run the cite checker against a brief where you already know which authorities are wrong. Count missed errors.
- Source linking: confirm that the linked passage actually supports the proposition.
- Confidentiality: check connector permissions and what each integration can read.
- Cost: ask about pricing, since none is published.
- Failure behavior: see what happens when a source is paywalled or missing; a tool should say "could not verify", not guess.
What this means for lawyers and builders
If you work in a firm, the near-term win is workflow, not magic: shared boards and matrices cut the time spent rebuilding trackers in spreadsheets. If you build legal tools, note the platform shift: Claude is turning into a distribution channel where partners publish artifacts. For teams running agents against client systems, an access-control layer matters; AgentBeam is the agent security platform from the explainx.ai team that stops AI agents before they take dangerous actions.
What is still unknown
Anthropic's page also does not say which Claude model powers the artifacts or how often the connected legal databases refresh, both of which affect how current a cite check can be.
- Pricing and regional availability.
- Independent accuracy measurements for the cite checker.
- How information barriers are technically enforced.
Related reading
Details are accurate as of October 28, 2026 and may change. This is not legal advice.