Objections, answered

In their own words.

Last revised 2026-05-20

A condensed answer to the questions a sharp legal mind raises about adopting Local Motion. Each entry leads with the actual phrasing we hear — not our marketing language — and replies in the vocabulary of the bar: rules, opinions, doctrines, by name.

If you find an objection here we haven't answered cleanly, tell us. We'd rather sharpen the answer than dodge the question.

What about Mata v. Avianca?

Mata v. Avianca · S.D.N.Y. 2023 · Castel, J. · Rule 11 The case is the cultural reference point for every legal-AI conversation since. Two lawyers cited cases that ChatGPT fabricated; Judge Castel imposed a $5,000 sanction under FRCP 11 and the matter became — properly — the floor for what counts as competent AI use in litigation. The Second Circuit's follow-up in Park v. Kim (2024) reinforced the principle.

Local Motion is, specifically, the architecture that answers this case. We don't tell the model "do not make up cases." We make it architecturally precluded for the model to emit an ID that isn't currently in the lawyer's library — the case_id, rule_id, and knowledge_id fields of the agent's tool schema are typed as JSON-Schema enums regenerated from your library on every reload. The model has no representation of "a case I'm thinking of that isn't in the list"; the only legal completions are members of the enum.

Said differently: the lawyers in Avianca used a chatbot in free-form mode and didn't verify before filing. With Local Motion, the model cannot fabricate, and every case shown in the panel has a verifiable canonical record (one-click to CourtListener — see citation verification). Independent verification before reliance remains your duty under Rule 3.3; we don't pretend otherwise.

What does ABA Formal Opinion 512 say about this?

ABA Formal Op. 512 · July 2024 Op. 512 addresses lawyer use of generative AI under the Model Rules across five surfaces: competence (Rule 1.1), confidentiality (1.6), communication with the client (1.4), candor (3.3) and supervision (5.1 / 5.3), plus fees (1.5). The throughline: the lawyer remains responsible for the output; the tool is the lawyer's instrument.

Local Motion's posture maps cleanly to that frame. The citation-lock supports competence (no fabricated authority to verify). The audio-stays-local-to-the-STT-provider-you-configure boundary supports confidentiality (see privilege below). The disclosure templates support candor — every paid plan ships pre-written language calibrated to the major state bars' published guidance. The review-and-approve library workflow supports supervision (you signed off on every entry the agent can cite).

Doesn't sending opposing counsel's words to OpenAI waive privilege?

FRE 502 · third-party vendor exception The short answer: the analysis is the same as for any third-party vendor providing services necessary for the rendition of legal services — e-discovery vendors, transcription services, Westlaw, Lexis, outside research providers. Disclosure to such a vendor under a confidentiality framework does not waive privilege when the disclosure is necessary to the representation and the vendor's terms support that posture.

Practically: only the text already spoken in open court reaches the LLM. Opposing counsel's argument, by definition, is on the record — it is not a privileged communication. Your own utterances tagged [ME] are similarly on the record, with the same observation. Your library — your work product — does reach the LLM as system context, which is why work-product doctrine matters.

For confidentiality of audio specifically: audio is captured locally and reaches the STT provider you configure (Groq, OpenAI Whisper, or any compatible endpoint) so it can be transcribed. The LLM never sees audio. To keep audio fully on-machine, configure stt.provider = "local" (whisper.cpp) or point at a self-hosted STT endpoint on your network. We document this honestly on the privacy policy and in Tech docs → Audio handling.

Isn't my library work product?

Hickman v. Taylor · 329 U.S. 495 (1947) Yes. The library is precisely "the attorney's mental impressions, conclusions, opinions, or legal theories" — the curated authorities you've decided are relevant, the holdings you've chosen to characterize a certain way, the recurring themes you've pre-tested. That makes the library protected work product under Hickman.

Local Motion's architecture is built for that posture. The library lives on your machine. The Document Ingest workflow proposes entries you review and approve — every entry has a name on it, namely yours. Local Motion does not aggregate libraries across customers, does not train any model on your library, and the agent only consults your library at trigger time (in BYO-keys mode, your library reaches your AI provider under the same vendor-disclosure framework as above).

Comment 8 to Rule 1.1 — am I obligated to use this kind of tool?

Model Rule 1.1 cmt. 8 · adopted 2012 Comment 8 imposes a duty to "keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology." That doesn't compel adoption of any specific tool. It does compel understanding.

The reframe that lands with most partners: it isn't "should we use AI." It's "we're already obligated to understand what AI is and isn't doing for the bar, and a research-and-preparation tool whose citations are architecturally constrained is closer to a binder with smart indexing than to ChatGPT in free-form mode." A reasonable lawyer evaluating tools under Comment 8 should evaluate this one.

Rule 3.3 — do I have to disclose to the court?

Model Rule 3.3 · candor to the tribunal · plus standing orders Disclosure rules vary by court, judge, and jurisdiction, and they're evolving. Several federal judges have issued standing orders requiring disclosure of generative-AI assistance — Judge Brantley Starr in N.D. Tex. was among the first. State bars in CA, FL, NJ, NY, and TX have published guidance; see the state ethics matrix for the rules-by-jurisdiction view.

Our position: disclosure should be the default. What you disclose is short — "I used a real-time research tool that retrieves authorities from a library I prepared in advance. The tool does not generate legal advice; every citation comes from materials I curated before this proceeding. The judgment is mine." Every paid plan ships pre-written language for a candor-to-tribunal certification, a notice of AI assistance, and a response to judicial inquiry. See the disclosure templates.

If the AI is treated like a non-lawyer assistant under Rule 5.3, who's responsible?

Model Rule 5.3 · responsibilities re non-lawyer assistance Op. 512 and most state bar opinions converge on treating generative AI as analogous to a non-lawyer assistant under Rule 5.3. That makes the supervising lawyer responsible for the work product the AI helped produce — the same standard as for a paralegal's research memo or a junior associate's draft.

Local Motion is built so that responsibility is straightforward to discharge. The agent never produces a final pleading; it surfaces prepared materials at the moment they're useful. Every authority shown comes from a library you reviewed and approved. Every advice card the agent surfaces is captured to a Post-hearing review window where you can rate and annotate it — that's the supervision trail.

Pro se users — isn't this UPL?

Model Rule 5.5 · unauthorized practice of law No. Local Motion does not represent anyone, does not advise anyone, and does not create an attorney–client relationship. For a pro se litigant it is a faster version of a Nolo self-help book or a court self-help center: research and preparation materials the litigant has chosen, organized for retrieval.

The architecture supports the UPL story. The tool does not generate citations — the litigant decides what's in the library before the hearing. The tool does not write briefs. The tool does not advise. It retrieves, faster than the litigant could on paper. That posture is consistent with how every state has carved out self-help resources from the practice of law, and it's the line we hold in the Terms.

What if the case in my library was overruled?

Shepardize / KeyCite is still your duty Two layers. First: every case entry can carry a superseded flag — overruled, limited, distinguished, abrogated, superseded by statute — and the panel renders flagged cases in red with a high-visibility badge so you don't read superseded authority into the record. See stale-law badge.

Second: every citation auto-checks against the CourtListener corpus at write-time and reports the precedential status. Non-Published / Errata / Separate / In-chambers statuses surface as warnings with a one-click flag-as-non-precedential. See citation verification.

What it does not do — and we say this plainly — is replace Shepardizing or KeyCite. The free CourtListener API doesn't expose a "still good law" boolean (that's what Shepard's and KeyCite are for, and they're not free). Local Motion catches typos and fabrications; it surfaces precedential-status warnings; it makes flagged cases impossible to miss in the panel. The citator-grade check before reliance remains your duty under Rule 3.3.

What if it mis-hears opposing counsel?

The most common failure mode the citation-lock can't address. Every advice card in the panel shows the captured they → "…" line with an inline "wrong?" link. One click copies the captured text into the Manual Input box (focused, caret at end). Correct the wording, hit Ctrl+Enter, the agent re-runs against your corrected version. See transcript correction.

Honest framing: transcription accuracy is bounded by the STT provider's quality on your audio. Whisper Large v3 Turbo (the default) handles courtroom audio well; judge microphones are often the weakest link. The "wrong?" affordance closes the loop fast enough that mis-transcription doesn't become mis-advice.

Could the session log become discoverable in malpractice?

FRCP 26 · work product doctrine · Hickman Session logs are off by default. When enabled, they are written only to your machine — never to our servers. You control retention. You control export. The format is JSON; nothing is locked away from you.

If a session log were ever sought in malpractice or motion practice, the standard work-product analysis applies — the log is the attorney's mental impressions about which authorities were considered relevant in a live proceeding. Whether that work product is opinion (absolutely protected) or fact (qualified protection) depends on the specific jurisdiction's treatment and the discovery posture. We don't claim immunity; we claim that the architectural choice to keep logs local-only, opt-in, and user-controlled is the right posture under modern work-product doctrine. The post-hearing review window exports a Markdown or JSON snapshot you can decide what to do with.

What if it crashes mid-hearing?

Plan for it. Every operational decision is shaped by this fear:

• The agent dispatcher is independent of the audio capture; if the agent fails, audio capture and transcription keep working — you can still read the transcript with Ctrl+T and decide on your own.
• Crash recovery on the post-hearing Review session is automatic; the in-memory record is persisted on every write to %APPDATA%\local-motion\sessions\active.json, atomic temp+rename, so a force-quit doesn't lose ratings or notes.
• Ctrl+Esc is panic-hide; the panel disappears instantly with no animation. Ctrl+\ brings it back. Useful when you suddenly need to screen-share and don't want the panel visible.
• The fallback is always your prep. Your binder, your notes, your outline. The tool makes you faster when it works; it doesn't make you helpless when it doesn't.

Isn't this just Harvey or CoCounsel or Spellbook?

Casetext to Thomson Reuters · $650M · 2023 No. Those products are async drafting tools — research memos, contract review, brief outlines — produced before or after a proceeding. Local Motion is real-time, in-hearing. Different surface, different latency budget, different failure modes.

The category leaders matter as comps for investors and as anchors for buyers' expectations, but they're not the same product. Harvey is in the asynchronous research-and-drafting market that Thomson Reuters validated when they paid $650M for Casetext / CoCounsel in 2023. Local Motion is in a much narrower lane — real-time motion practice, depositions, and oral argument — with the only architecturally hallucination-free posture in the category. That's a compliance moat that the async incumbents can't match without a ground-up rebuild of how their tools construct citations.

What's my Rule 11 exposure with this?

FRCP 11 · Mata v. Avianca Properly bounded. The Avianca sanction landed because the lawyers (a) cited cases that didn't exist and (b) didn't verify before filing. Local Motion eliminates (a) by architecture and reduces (b) by making CourtListener verification a one-click affordance at every write point.

That said: Rule 11 attaches to your signature, not to a tool. If you read aloud something the agent surfaces without confirming it's responsive to the actual record, you bear the consequences. The tool reduces the surface area; it does not eliminate the duty. Independent verification under Rule 3.3 and FRCP 11 is always yours.

Have a specific objection we haven't addressed? Tell us. We track every distinct concern we receive and either land the implementation, sharpen the answer, or — when it's genuinely unanswerable — say so plainly.