A company holding patents tied to two computer scientists has sued Anthropic in federal court, accusing the AI firm's Claude models of building on technology it never licensed. Flinn Artificial Intelligence Technologies filed the case Thursday in the U.S. District Court for the District of Delaware, alleging infringement of five patents covering attention-based neural-network training and processing.
According to Boursorama, the suit alleges that Anthropic's Claude large language models infringe the patents, and it seeks an unspecified amount of damages. The case is captioned Flinn Artificial Intelligence (AI) Technologies LLC v. Anthropic PBC and carries the docket number 1:26-cv-01454, per the same report.
Flinn AI Technologies says inventors Steven Flinn and Naomi Moneypenny filed the underlying machine-learning patents in 2015 and 2016. The company alleges Anthropic built its business on the reasoning, search, memory, coding, agentic, and multimodal capabilities described in those patents, the Boursorama report states.
Five Patents Named in the Filing
A separate account from Law360 lists the five patents at issue as Nos. 12,223,404, 12,327,166, 12,462,197, 12,462,198, and 12,511,580, each titled “Iterative attention-based neural network training and processing.” That detailed breakdown offers the clearest look yet at exactly what Flinn claims Anthropic copied.
Flinn also filed a parallel case against OpenAI in Delaware over the same five patents. The two complaints were filed the same day in Delaware federal court.
Patent Suits Against Anthropic Have Been Uncommon
Patent-infringement lawsuits against Anthropic have historically been rare compared with the wave of litigation over its use of copyrighted material to train its systems, Boursorama notes. That makes this week's filing a notable departure from the legal fights that have defined Anthropic's recent history, even as the underlying allegations remain unproven.
The case also lands amid a broader surge in generative-AI patent activity. A Patsnap analysis counted 29,906 published records filed against generative-model and training-data claim language through August 2026, with annual filings climbing from 1,121 in 2021 to 6,361 in 2024. Patsnap cautions that a roughly 18-month publication lag means its 2025 and 2026 figures likely understate real filing volume, and it found the ten largest patent holders account for just 22.8% of all records in scope, a sign the field remains fragmented rather than dominated by a handful of giants.
Not the First Delaware Fight Over Neural-Network Patents
This isn't the first time Delaware's federal court has hosted a neural-network patent dispute this year. In May 2026, five companies building AI transcription, speech-to-text, and customer-experience products, including AssemblyAI Inc., Cerence Inc., and NVIDIA Corp., were sued in the same district over alleged infringement of a different patent, No. 10,984,320, titled “Highly trainable neural network configuration,” according to Law360's earlier coverage.
No response from Anthropic, OpenAI, Flinn Artificial Intelligence Technologies, or the inventors was included in the available reporting, and the current procedural status of either Delaware case remains unclear. The allegations have not been tested in court, and neither the merits of the patent claims nor whether infringement actually occurred has been established.
