AI Infringement and Unfair Competition Response

These matters usually arrive as defence and offence at the same time. Model output may be alleged to infringe, training data may be alleged to be unlawfully sourced, and at the same moment a competitor's scraping and traffic diversion is eroding the company's own position. The first task is to separate whether a claim holds from whether the cost is controllable — the two do not share a test.

Defence and offence usually arrive together

These matters usually arrive as defence and offence at the same time. Model output may be alleged to infringe, training data may be alleged to be unlawfully sourced, and at the same moment a competitor’s scraping and traffic diversion is eroding the company’s own position. The first task is to separate whether a claim holds from whether the cost is controllable — the two do not share a test.

Three categories of claim

Copyright claims concentrate at two points: training data and output. The training stage involves acts of reproduction; the output stage involves a finding of substantial similarity. Judicial practice on the originality of AI-generated content is not yet uniform, and outcomes vary materially by case. Trademark and unfair competition claims concentrate on output that trades on another’s marks or goodwill. Personal information claims target the lawfulness of processing personal information contained in training data.

Defence routes

Fair-use defences in the training-data context still lack a uniform standard and have to be argued case by case, weighing the purpose of use and the effect on the original’s market. On the output side, the defence turns on showing that the generation process and the result do not amount to substantial similarity. Technological neutrality is not a standalone defence, but it can support an account of the actor’s state of mind. Defence strategy needs to be set early; there is limited room to change it later.

Offence: scraping and traffic diversion

Where a competitor systematically scrapes platform data or diverts traffic by technical means, the claim is usually brought under the Anti-Unfair Competition Law (2025 revision). The difficulty is evidence: fixing the technical traces of the scraping, proving the commercial value of the data, and quantifying loss or the competitor’s gain. Pre-litigation conduct preservation is effective where urgency exists, but the requirements on security and proof of urgency are high.

Linking platform complaints and litigation

Most AI products face platform complaints and judicial proceedings at the same time. A platform complaint is fast and cheap and can stop losses, but produces no res judicata; litigation is slow and can produce a final resolution. The order in which the two run affects the scope of disclosure and the negotiating position, and should be decided before either begins.

Controlling cost

The spend on these matters frequently exceeds the amount in dispute. A workable approach: complete an infringement comparison and liability assessment before deciding whether to proceed; use a standard evidence template across a batch of claims to lower per-case cost; and where a commercial relationship remains, substitute licensing or settlement for litigation. Cost should be judged alongside commercial objectives rather than left to the procedure.

Related services: IP Protection, Data Compliance and Privacy, AI Content Labelling and Content Safety, Commercial Contracts.