DCT
2:26-cv-00774
Many Worlds 2T Innovations LLC v. OpenAI OpCo LLC
Key Events
Complaint
Table of Contents
complaint Intelligence
I. Executive Summary and Procedural Information
- Parties & Counsel:
- Plaintiff: Many Worlds 2T Innovations LLC (Texas)
- Defendant: OpenAI OpCo, LLC (Delaware)
- Plaintiff's Counsel: Cadwell Thomas LLP
- Case Identification: 2:26-cv-00774, E.D. Tex., 08/31/2026
- Venue Allegations: Venue is alleged to be proper in the Eastern District of Texas because Defendant OpenAI has committed acts of infringement in the District and maintains a regular and established place of business at an artificial intelligence data center campus in Denton, Texas.
- Core Dispute: Plaintiff alleges that Defendant's artificial intelligence products, including ChatGPT and related APIs, infringe five U.S. patents related to contextual search, inferential-based communication, and adaptive content discovery systems.
- Technical Context: The technology at issue involves methods for improving computer-implemented recommendation and search systems by moving beyond simple keyword matching or collaborative filtering to incorporate semantic analysis, user behavior, and explicit user controls.
- Key Procedural History: The complaint states that the asserted patents claim priority to a series of provisional applications dating back to March 2011 and arise from a continuous chain of co-pending applications. The complaint also makes extensive pre-emptive arguments regarding the patent eligibility of the claims under 35 U.S.C. § 101, suggesting that Plaintiff anticipates and is prepared to litigate a subject-matter eligibility challenge under the framework established by Alice Corp. v. CLS Bank Int'l.
Case Timeline
| Date | Event |
|---|---|
| 2011-03-29 | Earliest Priority Date for all Asserted Patents ('742, '202, '388, '603, '433 Patents) |
| 2014-03-18 | U.S. Patent No. 8,676,742 Issues |
| 2014-09-23 | U.S. Patent No. 8,843,433 Issues |
| 2020-06-30 | U.S. Patent No. 10,699,202 Issues |
| 2025-05-13 | U.S. Patent No. 12,299,603 Issues |
| 2025-05-20 | U.S. Patent No. 12,307,388 Issues |
| 2025-09-26 | OpenAI registers as a foreign entity in Texas |
| 2025-11-24 | ChatGPT Shopping Research launched |
| 2025-12-31 | Denton Campus becomes operational (approximate date) |
| 2026-08-26 | Assistants API File Search discontinued |
| 2026-08-31 | Complaint Filed |
II. Technology and Patent(s)-in-Suit Analysis
U.S. Patent No. 8,676,742 - "Contextual Scope-Based Discovery Systems"
- Patent Identification: U.S. Patent No. 8,676,742, issued on March 18, 2014 (the '742 Patent).
- The Invention Explained:
- Problem Addressed: The patent addresses the deficiencies of prior recommendation engines that relied predominantly on collaborative filtering (Comp. ¶26). These systems were "content-agnostic," treating content as opaque identifiers, which created "cold-start" problems for new content and provided "black-box" recommendations without explanation, reducing user trust and engagement Compl. ¶26
- The Patented Solution: The invention proposes a "fuzzy network-based structure" where content and topic "objects" are linked by weighted "affinity relationships" Compl. ¶1 Compl. ¶29 As illustrated in Figure 3 of the complaint, this network uses numerical weights to indicate relationship strength between different objects Compl. p. 8 A "contextual scope function" establishes a "contextual neighborhood" within this network, and a "discovery function" generates recommendations based on that neighborhood and inferences from user behaviors Compl. ¶32 '742 Patent, col. 2:20-30 This allows for more nuanced, probabilistic recommendations that account for both content characteristics and user activity Compl. ¶29
- Technical Importance: The invention represents a technological improvement over purely content-blind collaborative filtering by integrating content analysis into the recommendation process through a specific data structure.
- Key Claims at a Glance: The complaint asserts at least independent claims 1 and 8 Compl. ¶54
- Independent Claim 1 (Method):
- invoking a contextual scope function...that establishes a contextual neighborhood within a computer-implemented fuzzy network-based structure;
- receiving a recommendation...generated by a computer-implemented discovery function that generates the recommendation in accordance with the contextual neighborhood and an inference from a plurality of usage behaviors; and
- receiving the recommendation, wherein the...contextual neighborhood is based on a selected object.
- Independent Claim 8 (System):
- a contextual scope function...that establishes a contextual neighborhood within a computer-implemented fuzzy network-based structure;
- a contextual neighborhood scoping function that establishes the scope of the contextual neighborhood at a level set by a user; and
- a computer-implemented discovery function that generates a recommendation in accordance with the contextual neighborhood and an inference from a plurality of usage behaviors.
- Independent Claim 1 (Method):
U.S. Patent No. 10,699,202 - "Inferential-based Communications Method and System"
- Patent Identification: U.S. Patent No. 10,699,202, issued on June 30, 2020 (the '202 Patent).
- The Invention Explained:
- Problem Addressed: The patent seeks to solve the "long-standing and pervasive problem" of users not having context for recommendations, which left them "baffled as to why a recommendation was made" Compl. ¶33
- The Patented Solution: The invention describes a method for automatically generating a "communication" (i.e., an explanation). The system automatically accesses two sets of data: a "first plurality of values" derived from analyzing text associated with a content object (e.g., content relevancy scores), and a "second plurality of values" derived from user behaviors (e.g., user-topic interest levels) Compl. ¶34 '202 Patent, col. 61:1-15 It then generates a communication by selecting words "in accordance with the first plurality of values, the second plurality of values, and one or more syntactical rules" '202 Patent, claim 1 This computationally constructs an explanation from analyzed data rather than using a pre-determined template Compl. ¶34
- Technical Importance: This approach solves the prior art's "black box" problem by creating a specific technical pipeline for generating dynamic, data-driven explanations for recommendations.
- Key Claims at a Glance: The complaint asserts at least claims 1, 8, 15, and 20 Compl. ¶73
- Independent Claim 1 (Method):
- accessing automatically a first plurality of values, determined by automatically analyzing text associated with a content object, based on an inferred degree of relevancy;
- accessing automatically a second plurality of values, corresponding to topics and based on an inference from user behaviors; and
- generating automatically a communication for a user, comprising a plurality of words selected in accordance with the first values, the second values, and one or more syntactical rules.
- Independent Claim 1 (Method):
U.S. Patent No. 12,307,388 - "Probabilistically Tunable Conversational Method and System"
- Patent Identification: U.S. Patent No. 12,307,388, issued May 20, 2025 (the '388 Patent).
- Technology Synopsis: The patent addresses the problem of users having little to no control over the serendipity of recommendations Compl. ¶35 It claims a method that provides a "serendipity tuning control" to a user, applies the user's setting to a "probabilistic selection algorithm," and generates a communication based on this tuned algorithm Compl. ¶36
- Asserted Claims: At least claims 1, 8, and 15 Compl. ¶90
- Accused Features: The accused features include the "temperature" and "top_p" parameters in the OpenAI Chat Completions API, which are alleged to function as the claimed "serendipity tuning control" Compl. ¶90
U.S. Patent No. 12,299,603 - "Vector-Based Search Method and System"
- Patent Identification: U.S. Patent No. 12,299,603, issued May 13, 2025 (the '603 Patent).
- Technology Synopsis: The patent addresses search systems that retrieve by matching words rather than by comparing meaning Compl. ¶37 It claims a method where neural networks generate "affinity vectors" for both stored content and a user's search input; the query vector is then compared against the stored vectors using a "mathematical-based vector comparison algorithm" to retrieve content Compl. ¶38
- Asserted Claims: At least claims 1, 8, and 15 Compl. ¶107
- Accused Features: The accused features include OpenAI's retrieval stack, which allegedly embeds content and queries into a vector space, compares them using cosine similarity, and returns the corresponding content Compl. ¶108
U.S. Patent No. 8,843,433 - "Integrated Search and Adaptive Discovery System and Method"
- Patent Identification: U.S. Patent No. 8,843,433, issued September 23, 2014 (the '433 Patent).
- Technology Synopsis: The patent addresses the problem of prior recommenders using either user behavior data or content analysis, but not both within a single function Compl. ¶39 The invention claims a "two-input architecture" where a single recommendation function generates a recommendation "in accordance with" both a first set of values based on user behavior and a second set of values based on content analysis Compl. ¶40
- Asserted Claims: At least claims 1, 8, and 17 Compl. ¶124
- Accused Features: The accused features include OpenAI's "retrieval-augmented recommendation platform," which is alleged to combine user interest inferences from ChatGPT Memory with relevance scores from vector retrieval to generate recommendations Compl. ¶124 Compl. ¶125
III. The Accused Instrumentality
Product Identification
The Accused Instrumentalities are a broad suite of OpenAI's artificial intelligence products and services, including but not limited to ChatGPT (all tiers), ChatGPT Memory, ChatGPT Pulse, ChatGPT Shopping Research, OpenAI Vector Stores, the Vector Store Search endpoint, various APIs (Chat Completions, Completions, Embeddings), and the text-embedding-3 model family Compl. ¶46
Functionality and Market Context
The complaint alleges that the Accused Instrumentalities implement "dual-analysis architectures" that use separate neural networks to process user behavioral data and content characteristics Compl. ¶50 This process generates "weighted embeddings" that are then combined to produce content for a user Compl. ¶50 Specifically, OpenAI's Vector Stores are alleged to implement a structure of nodes (file chunks) joined by "graded affinities" in the form of cosine similarities between their embedding vectors Compl. ¶50 The complaint describes the ChatGPT with Memory feature as building persistent user interest profiles from conversational behaviors Compl. ¶47 The complaint's Figure 3, a diagram from the '742 Patent, is presented as an example of the "fuzzy network" architecture that Plaintiff alleges is mirrored by OpenAI's systems Compl. ¶27 Compl. p. 8
IV. Analysis of Infringement Allegations
U.S. Patent No. 8,676,742 Infringement Allegations
| Claim Element (from Independent Claim 8) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| a contextual scope function executed on a processor-based computing device that establishes a contextual neighborhood within a computer-implemented fuzzy network-based structure; | The vector-store retrieval index over embedded content, where a search returns a subset of objects with a specified degree of relatedness, each with a numerical score. | ¶54 | col. 2:20-25 |
| a contextual neighborhood scoping function that establishes the scope of the contextual neighborhood at a level set by a user; and | OpenAI's file-search ranking controls, such as the score_threshold on a 0-to-1 scale, the maximum-results setting, and other parameters that allow a user to configure the breadth of the search. |
¶54 | col. 21:50-55 |
| a computer-implemented discovery function that generates a recommendation in accordance with the contextual neighborhood and an inference from a plurality of usage behaviors. | The recommendation functionality of ChatGPT Memory and ChatGPT Shopping Research, which are alleged to draw on a user's behaviors over time to generate recommendations. | ¶54 | col. 2:25-30 |
U.S. Patent No. 10,699,202 Infringement Allegations
| Claim Element (from Independent Claim 1) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| accessing automatically a first plurality of values, wherein each of the first plurality of values is determined by automatically analyzing text that is associated with a content object... | The per-chunk relevance scores that OpenAI's file-search documentation describes as being returned with each retrieved chunk, along with a score threshold. | ¶73 | col. 10:1-13 |
| accessing automatically a second plurality of values, wherein each of the second plurality of values corresponds to one or more topics... and wherein a magnitude of each... is based upon an automatically determined inference from one or more user behaviors; and | The per-user, topic-organized store of preferences, interests, and goals that ChatGPT Memory allegedly builds automatically from a user's chat history. | ¶73 | col. 10:19-34 |
| generating automatically a communication for delivery to a user... wherein the plurality of words are selected in accordance with the first plurality of values, the second plurality of values, and one or more syntactical rules. | OpenAI's Structured Outputs feature, which allegedly converts a JSON Schema into a context-free grammar to enforce token-by-token sampling and constrain the output. | ¶73 | col. 57:15-24 |
Identified Points of Contention
- Scope Questions: A central dispute may concern whether OpenAI's use of neural network-generated embeddings and vector databases constitutes a "computer-implemented fuzzy network-based structure" as recited in the '742 Patent. The patent describes this structure using terms like "topic objects" and "content objects" with "relationships" and "affinities" Compl. ¶¶27-29 The infringement analysis will likely turn on whether OpenAI's architecture of file chunks, embedding vectors, and cosine similarity scores Compl. ¶50 can be read to fall within the patent's claimed structure.
- Technical Questions: For the '742 Patent, a question arises as to whether user-configurable API parameters like a score threshold or a maximum number of results Compl. ¶54 perform the same function as the claimed "contextual neighborhood scoping function that establishes the scope... at a level set by a user." For the '202 Patent, a key technical question will be whether the accused generative AI systems "select" words in accordance with the three claimed inputs (content values, user behavior values, and syntactical rules), or whether they operate in a more integrated, holistic manner that does not map directly onto the patent's specific, ordered computational pipeline.
V. Key Claim Terms for Construction
U.S. Patent No. 8,676,742
- The Term: "computer-implemented fuzzy network-based structure"
- Context and Importance: This term is the foundational data structure of the asserted claims of the '742 Patent. The outcome of the infringement analysis may depend on whether OpenAI's vector embedding and retrieval system is construed to be a "fuzzy network-based structure."
- Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The patent specification defines a "fuzzy network" as "a computer-implemented plurality of nodes, with relationships among the nodes that have affinities that are by degree" '742 Patent, col. 4:1-4 Plaintiff may argue this broad definition covers any system with items (nodes) and graded connections (affinities), such as OpenAI's vector embeddings and cosine similarity scores.
- Evidence for a Narrower Interpretation: The specification's detailed description and figures, such as Figure 3 Compl. p. 8, depict a specific architecture of "topic objects" and "content objects" '742 Patent, col. 7:5-10 Defendant may argue the term should be limited to this disclosed embodiment, where "topic objects" are "labels" that do not contain pointers to information, a characteristic that may not be present in OpenAI's system Compl. ¶28 '742 Patent, col. 6:64-7:2
U.S. Patent No. 10,699,202
- The Term: "words are selected in accordance with the first plurality of values, the second plurality of values, and one or more syntactical rules"
- Context and Importance: This limitation in claim 1 of the '202 Patent defines the core mechanism for generating the "communication." Practitioners may focus on this term because the infringement case depends on mapping the complex, probabilistic process of a large language model onto this specific three-part constraint.
- Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The claim language is functional. Plaintiff may argue that it covers any computational process that generates text where the word choice is constrained by content relevance (first values), user profile (second values), and grammar (syntactical rules), regardless of the specific implementation.
- Evidence for a Narrower Interpretation: The claim recites a specific order of operations: accessing two distinct sets of values and then selecting words based on them. Defendant may argue that this requires a discrete, multi-step algorithm, rather than the integrated, token-by-token generation process of a transformer-based language model, where such constraints are applied more holistically during sampling from a vocabulary distribution.
VI. Other Allegations
Indirect Infringement
The complaint alleges that OpenAI induced infringement of the method claims by providing its AI products to end users and developers. The alleged inducing acts include providing instructions on how to use the infringing features, marketing and promoting these capabilities, and publishing API documentation, guides, and tutorials that instruct users on how to perform the claimed methods Compl. ¶60 Compl. ¶77 Compl. ¶93 Compl. ¶111 Compl. ¶128
Willful Infringement
The complaint alleges that OpenAI has had knowledge of the patents at least as of the filing of the complaint. It asserts that any continued infringement after this notice is "willful, deliberate, and in reckless disregard" of Plaintiff's rights. The complaint explicitly reserves the right to allege pre-suit willfulness if evidence of prior knowledge is found during discovery Compl. ¶67 Compl. ¶84 Compl. ¶101 Compl. ¶118 Compl. ¶134
VII. Analyst's Conclusion: Key Questions for the Case
- A core issue will be one of definitional scope and technological evolution: Can the term "fuzzy network-based structure", rooted in a 2011-era paradigm of explicit, graph-like objects and relationships, be construed to cover a modern AI architecture based on high-dimensional vector embeddings and cosine similarity? The court's interpretation of this foundational term will significantly impact the infringement analysis for several of the asserted patents.
- A second central question will be one of patent eligibility: The complaint makes extensive arguments that the patents claim specific improvements to computer functionality, not abstract ideas. The case will likely feature a significant battle under 35 U.S.C. § 101 over whether the claims are directed to a specific, concrete technical solution (per Enfish and McRO) or to the abstract ideas of generating recommendations, searching for information, and creating explanations.
- A key evidentiary question will be one of functional mapping: Does the operation of OpenAI's APIs and generative models perform the specific functions recited in the claims? For instance, does adjusting a "temperature" parameter in an API call equate to a "serendipity tuning control," and does a large language model's generative process "select words" in accordance with the distinct sets of values as required by claim 1 of the '202 patent? This will require a detailed technical comparison between the patented methods and the accused systems' actual operations.
Analysis metadata