AI in Practice: Prosecution and Litigation Strategies for Leveraging AI

Kilpatrick's Alton Absher, Dan Englander, Charles Gray, and Charlie Henn, joined by Jason Chang, AVP, Senior Legal Counsel at AT&T, presented “AI in Practice: Prosecution and Litigation Strategies for Leveraging AI” at Kilpatrick's 2026 Intellectual Property Seminar Series in Atlanta on September 16, 2026. The session opened with a live audience survey repeating questions first asked a year earlier, set out the professional responsibility framework governing a lawyer's use of Generative AI, and then ran recorded demonstrations of agentic workflows from invention intake and quality assurance through portfolio analysis and the assertion lifecycle. It turned last to the trademark docket, where the same problems arrive on different files.

The panel provides these key takeaways:

1. The duty does not change because the tool is new, but which part of it bites changes with the step. The panel ran one framework, competence and diligence alongside supervision and accountability, and keyed each demonstration to the part governing that step. Ideation output is a candidate list, not a filing decision, and 37 CFR 11.18(b)(2) requires a contention presented to the Office to rest on an inquiry reasonable under the circumstances. Quality assurance runs on competence. Triaging an inbound demand runs on supervision, where Florida Bar Ethics Opinion 24-1 applies Model Rule 5.3 to Generative AI by analogy rather than by its terms. A control written for one step does not discharge what the next one triggers.

2. Verification failure is no longer only a court problem: in 2026 the USPTO publicly reprimanded a registered practitioner under its own diligence rule, 37 CFR 11.103. That rule mirrors ABA Model Rule 1.3 word for word. Neither text mentions technology, so the pre-signature review duty sits in the reasonable inquiry 37 CFR 11.18(b) certifies and in competence, with diligence carrying it as applied. The filing there was a claim construction chart in district court rather than a paper before the Office, and the citations were to the patent's own intrinsic record, a pitfall the order warns “may also extend to the intrinsic evidence of patent and trademark applications and their file wrappers.” The same order found candor violations, which is the exposure that opens after the error rather than before it.

3. Routing portfolio material through any tool is a confidentiality decision before it is a technology decision. The portfolio demonstration was built around three things the audience was told to watch for: claims mapped to products and technical areas, coverage gaps and white space flagged, and ranked recommendations for the team. What has to go in to get that out is client information. Under 37 CFR 11.106(d) a practitioner must make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client, which reaches considerably more than what is secret. Settle where that information goes before the first upload.

4. One demonstration was built on the output of the one before it, and that single hand-off is the part worth generalizing from. The assertion demonstration did not begin from a blank page. It ran on the portfolio analysis from the prior demonstration: the portfolio read went in, and a claim chart, an invalidity analysis and an assertion roadmap came out. That is the only hand-off the six demonstrations document, and it is the one that tells an in-house team where to start, because a step fed by the one ahead of it is worth more than the same step run alone. Pilot first wherever someone is currently rebuilding by hand what an earlier step already produced.

5. Accountability has a moment as well as an owner, and the moment is not the prompt. Fed. R. Civ. P. 11(b) and 37 CFR 11.18(b) attach on presenting a paper to a court or to the Office, by signing, filing, submitting, or later advocating it. Neither reaches the draft demand letter the assertion demonstration ends on; the claim chart behind it is what the certification is measured against once a complaint is filed. The bench-preparation set from the other demonstration is preparation rather than authority, which the deck conditions on verifying every cite. The Ninth Circuit's published order in Lnu v. Blanche, No. 24-4790 (9th Cir. June 3, 2026), locates the violation “at the point of signing and filing.

6. The trademark docket raises the same problems, and the panel presented every response conditionally, as what the work could look like rather than as tooling that exists today. The problem set was long weekly watch notice lists in which few entries matter, office action dockets carrying deadlines without end, and repetitive enforcement against many infringers with a backlog of partly finished settlements. For each the panel walked a before and after: ingestion and deadline sorting, an investigation step used as appropriate, an attorney-built enforcement framework queried against historical enforcement decisions, review against a vault of best-of responses and templates, a running log, and a human in the loop on every one. What the design turns on is that record of decisions already made, ahead of a recommended action and an escalation path.