AI at the ITC in 2026: Section 337 Fault Lines for Training Data, Software, and AI-First Devices

Litigation AI Monitor - February 2026: AI, IP, and Section 337 at the ITC

AI is no longer experimental. It's embedded in chips, phones, cars, and cloud stacks. With that comes a predictable next step: more Section 337 complaints at the US International Trade Commission (ITC). This piece maps where AI disputes are headed, how Section 337 fits, and where jurisdiction may stall.

Why think about AI-related ITC cases now?

US companies are pouring capital into AI R&D and want fast, market-wide remedies when competitors import infringing tech. Section 337 offers exactly that-rapid schedules and exclusion orders with teeth. As AI features ship in physical products, the ITC becomes an obvious forum.

The ITC's role in protecting IP

The ITC adjudicates unfair import practices. Most cases are about patents, copyrights, and trademarks, but trade secrets show up too. If a violation is found, the Commission can issue exclusion and cease-and-desist orders that shut off access to the US market for the accused articles.

Elements of a Section 337 violation (federal IP)

• Importation, sale for importation, or sale after importation of an article.
• Infringement under the relevant federal IP law (patent, copyright, trademark).
• A domestic industry that meets economic and technical prongs.

Importation can be minimal-one unit, a sample, a trade show demo, or material used in trials. "Sale for importation" covers deals made with intent or knowledge that the article will enter the US.

Infringement is assessed under standard federal doctrines, including direct and indirect (inducement, contributory) theories. The domestic industry requirement asks for significant US investment (plant, labor, R&D, or licensing) and proof that a domestic article practices at least one claim or right at issue.

The ITC's track record with new tech

From memory chips and batteries to smartphones and medical devices, the ITC stays where the commercial action is. The draw is speed and strong remedies. Expect AI-related cases to follow the same path as demand spikes and imports rise.

Where AI is most likely to trigger ITC disputes

• AI training data and model training pipelines.
• AI hardware (accelerators, edge chips, sensors, smart modules).
• Software using AI (on-device and cloud-connected features).
• AI-first products (robotic systems, vision-enabled devices, wearables).

Hardware and AI-first products fit neatly at the ITC. Data pipelines and software delivered electronically raise jurisdictional questions that respondents will press.

2025 signals: what moved the needle

Consumer and enterprise devices shipped with on-device AI, and several Section 337 complaints followed. Examples filed in 2025 include:

• AI chips for autonomous systems (337-TA-1450; 337-TA-1443).
• Wearable EEG devices with AI algorithms (337-TA-1458).
• Wi-Fi routers using AI for network management (337-TA-1454).
• Active electrical cables for AI data backends (337-TA-1446).

Expect more filings as AI becomes a headline feature in phones, laptops, TVs, cars, and wearables.

Unique legal wrinkles for AI at the ITC

For hardware and AI-first devices, nothing unusual-these are tangible articles, long familiar at the ITC. The friction point is intangibles: training data and software delivered electronically. That is where ClearCorrect limits jurisdiction and where complainants will need strategy.

Training data: the ClearCorrect constraint

In ClearCorrect, the Federal Circuit held that electronically transmitted digital data is not an "article" under Section 337. That narrows the path for complaints centered on digital-only imports like model weights or training sets sent over the wire. Complaints hinging on data alone are likely to face threshold challenges.

ClearCorrect in brief

The case involved digital 3D dental models transmitted from abroad to the US for making aligners. The Commission found a violation; the Federal Circuit reversed, holding that "articles" means material things, not digital transmissions. The result: the ITC cannot regulate pure digital imports under Section 337.

What about copyrights and GenAI output?

In 2025, a district court in Bartz v. Anthropic found that training on copyrighted books qualified as fair use, while criticizing the mass copying and storage of pirated works and leaving room for liability tied to outputs. Even so, ClearCorrect means digitally transmitted outputs that infringe copyrights fall outside the ITC's reach. For physical AI-first products trained on copyrighted works, liability at the ITC will likely depend on whether the device reproduces protected expression, not merely that training used protected material.

Does Loper Bright change ClearCorrect?

Unlikely. The Supreme Court ended Chevron deference in Loper Bright, but ClearCorrect did not turn on agency deference in a way that helps complainants here. The Federal Circuit grounded its decision in statutory text, usage, and practical enforceability. Expect ClearCorrect to stand unless Congress amends the statute.

For reference: 19 U.S.C. § 1337 and the Court's opinion in Loper Bright.

Software using AI: where jurisdiction can still hold

AI software delivered electronically, without a tied physical article, is outside the ITC under ClearCorrect. But there's a play under Suprema: target the imported hardware and prove inducement where software is added or developed post-importation in the US. That theory is alive, and it's practical in sectors where hardware ships first and gets AI features installed later.

Suprema in brief

Suprema imported scanners. US partner Mentalix developed software domestically using an SDK and documentation from Suprema. The ITC found direct infringement by Mentalix and induced infringement by Suprema, and the Federal Circuit (en banc) affirmed. Key point: importing hardware while enabling or encouraging domestic software integration that completes the infringement can satisfy Section 337.

Applying Suprema to AI software

Think camera and LiDAR modules imported without ML software, then US dealers install lane keeping or sign recognition after the fact with the importer's help. A patent owner could pursue the module maker for inducement at the ITC. ClearCorrect doesn't undercut this because the accused "article" is the imported hardware, not a digital transmission.

Why inducement theories may be tested

After Loper Bright, respondents may ask courts to re-examine Federal Circuit decisions that leaned on Chevron, including Suprema. That said, recent decisions (for example, Lashify on domestic industry) show courts are not uniformly narrowing ITC reach. The Commission has also noted that ending Chevron does not automatically undo its Section 337 interpretations. Still, expect briefing on how much involvement in importation or post-import steps is sufficient for liability.

Jurisdictional playbook: complainants

• Anchor claims to tangible imports. Pair software claims with imported hardware, components, or physical media.
• Build inducement evidence: SDKs, manuals, reference designs, training, integration support, and internal communications showing knowledge and intent.
• Prove domestic industry early. Tie US investments to practicing products and map claim charts for the technical prong.
• Anticipate ClearCorrect defenses. Avoid complaints that rely solely on data or cloud transmissions.
• Select remedies with intent. Request limited or general exclusion orders and cease-and-desist orders based on the distribution model.

Jurisdictional playbook: respondents

• Press ClearCorrect on any digital-only theory (training data, APIs, cloud-only outputs).
• Break the inducement chain. Limit post-import support, strip out enablement materials, and document non-infringing uses.
• Contest domestic industry. Scrutinize allocation of R&D, headcount, and licensing spend; test whether the asserted product truly practices a claim.
• Challenge "sale for importation." Attack evidence of intent or knowledge that units would enter the US.
• Preserve Loper Bright arguments to revisit Chevron-era interpretations where helpful.

Practical scenarios to expect in 2026

• Phones and laptops with on-device AI accelerators: patent disputes over NPU/AI block architectures and on-device inference features.
• Vehicles with ADAS features: inducement theories where imported sensors are paired with domestically installed ML stacks.
• Wearables and health devices: method claims tied to signal processing and AI diagnosis/triage algorithms.
• Edge AI modules and gateways: trade secret claims over model weights, training pipelines, and deployment configs where hardware crosses the border.

Domestic industry: getting the record right for AI

• Allocate R&D to practicing products, not just platform research; make the link explicit.
• Track licensing revenue and engineering time tied to features practicing asserted claims.
• For software-heavy products, document the physical article that practices the claims (e.g., device + on-device model + feature path).

Bottom line

Hardware and AI-first devices are squarely within the ITC's reach. Pure digital imports are not. Use Suprema to connect imported hardware with post-import AI software where facts support inducement. Expect respondents to test those boundaries post-Loper Bright, but plan your record so the Commission has what it needs to issue exclusionary relief.