Intellectual Property

Anthropic and OpenAI Sued: New Neural Network Patent Risks for AI Developers

2026-11-18 · 12 min read · MeshLaw Newsroom

Source news: "Anthropic, OpenAI sued for allegedly infringing inventors' neural network patents" (Reuters) · Search original The following is original commentary written by AI based on facts verified from 3 real news reports (not a translation or copy of the original). See sources at the end.

Two patent infringement suits filed in the Delaware federal court against Anthropic and OpenAI have brought attention-based neural network patents into the spotlight, signaling a shift in how intellectual property risks are assessed for major AI developers. With five patents originally filed in 2015 and 2016 but only granted in 2025 now being asserted against flagship models like Claude and ChatGPT, in-house counsel must urgently re-evaluate their defensive strategies against legacy IP claims that may have previously flown under the radar.

The Sudden Surge in AI Patent Litigation

The legal landscape for artificial intelligence shifted dramatically in November 2026 when Flynn Artificial Intelligence Technologies LLC filed two separate patent infringement lawsuits against industry giants Anthropic and OpenAI. Filed in the U.S. District Court for the District of Delaware, these cases (1:26-cv-01454 against Anthropic and 1:26-cv-01455 against OpenAI) mark a significant escalation in litigation targeting foundational AI technologies. The plaintiff, a Texas-based patent holding company, alleges that Anthropic’s Claude models and OpenAI’s ChatGPT infringe upon five patents related to attention-based neural networks. Neither company has provided an immediate response to requests for comment regarding the filings.

What makes these suits particularly notable is the timing of the patent grants relative to the underlying technology. The patents in question, jointly invented by Stephen Flynn and Naomi Munipenny, were originally filed in 2015 and 2016 but were not granted until 2025. This decade-long gap between filing and issuance means that legacy intellectual property rights are now being asserted against modern AI systems that have matured significantly since the original applications. The sudden availability of these granted patents has created a new vector for liability, allowing patent holders to leverage older foundational concepts against current market leaders.

Key contextual factors driving this surge in litigation include:

  • Delayed Patent Grants: The patents were filed in 2015–2016 but only issued in 2025, creating a window where modern AI development proceeded without clear visibility into these specific IP rights.
  • Targeting Core Architecture: The claims focus on attention-based neural networks, a fundamental component of large language models like Claude and ChatGPT, rather than niche applications.
  • Strategic Venue: The choice of the District of Delaware suggests a strategic approach to jurisdiction, often favored for its experience with complex patent cases.
  • Patent Holder Profile: Flynn Artificial Intelligence Technologies LLC operates as a patent holding company, indicating a business model focused on monetizing intellectual property through enforcement rather than product development.

Anatomy of the Attention-Based Patent Claims

The infringement allegations center on five patents originally filed between 2015 and 2016, which were finally granted in 2025. These patents, jointly invented by Steven Flynn and Naomi Munipenny, cover attention-based neural network technologies. The timing of the grants is critical to the current legal landscape, as the patents had been pending for nearly a decade before issuance, allowing the inventors to potentially capture value from the rapid expansion of large language models that emerged years after the initial filings.

Flynn Artificial Intelligence Technologies LLC, a Texas-based patent holding company, filed the lawsuits in the U.S. District Court for the District of Delaware on November 12, 2026. The complaints specifically target Anthropic’s Claude models and OpenAI’s ChatGPT, alleging that these systems infringe upon the five granted patents. The cases are docketed as 1:26-cv-01454 against Anthropic and 1:26-cv-01455 against OpenAI, with the plaintiff seeking damages for the alleged unauthorized use of the patented attention mechanisms.

Key details regarding the patent portfolio include:

  • Inventors: Steven Flynn and Naomi Munipenny are listed as the joint inventors of the technology.
  • Filing Timeline: The patent applications were submitted in 2015 and 2016, predating the widespread commercial adoption of transformer-based architectures.
  • Grant Date: All five patents were officially granted in 2025, creating a narrow window for enforcement against current AI products.
  • Targeted Products: The litigation explicitly identifies Anthropic’s Claude and OpenAI’s ChatGPT as the infringing implementations of the claimed attention-based neural network methods.

Why Legacy Patents Pose a Modern Threat

The Gap Between Filing and Enforcement

The core of the current legal tension lies in the significant time lag between patent filing and enforcement. The patents at the center of the Flynn Artificial Intelligence Technologies LLC lawsuits were filed in 2015 and 2016, years before the transformer architecture became the dominant paradigm for large language models. These patents, invented by Steven Flynn and Naomi Moneypenney, were not granted until 2025, creating a nearly decade-long period where the underlying technology evolved rapidly while the patent applications sat in examination. This timeline allows patent holders to secure rights on foundational concepts that were in their infancy at the time of filing but have since become ubiquitous in modern AI development.

By the time these patents were registered in 2025, the industry had already standardized around attention-based mechanisms similar to those described in the original filings. This creates a scenario where companies like Anthropic and OpenAI, which built their flagship products—Claude and ChatGPT—on architectures that were considered cutting-edge but not necessarily novel in the context of the 2015-2016 prior art landscape, now face infringement claims. The patents were filed when neural network research was largely experimental, yet they are being enforced against mature, commercially deployed systems that rely on the very mechanisms those early filings anticipated.

Strategic Timing and Litigation Leverage

The strategic value of these legacy patents is amplified by their timing relative to the AI boom. Flynn Artificial Intelligence Technologies LLC, a Texas-based patent holding company, filed suit in the U.S. District Court for the District of Delaware in November 2026, targeting the two most prominent AI developers. The choice of venue and the specific targeting of market leaders suggest a strategy designed to maximize settlement leverage rather than purely seek injunctive relief.

Key factors driving this threat include:

  • Retroactive Application: Patents filed in 2015/2016 are being applied to products that did not exist in their current form at the time of filing, raising questions about the scope of the claims versus the state of the art.
  • Standardization of Architecture: The widespread adoption of attention-based models means that many AI developers may inadvertently infringe on broad, early-stage patents that were granted years later.
  • High-Profile Targets: By suing Anthropic (Case No. 1:26-cv-01454) and OpenAI (Case No. 1:26-cv-01455), the plaintiff is leveraging the high visibility and financial resources of these companies to set a precedent.
  • Delayed Grant Process: The 2025 grant date means these patents were not publicly enforceable during the critical growth phase of the LLM market, allowing developers to build without awareness of the impending liability.

Immediate Liability Exposure for AI Developers

Financial and Operational Exposure for AI Developers

The lawsuit filed by Flynn Artificial Intelligence Technologies LLC against Anthropic and OpenAI in the U.S. District Court for the District of Delaware introduces immediate financial uncertainty for developers relying on attention-based architectures. By targeting five patents originally filed in 2015 and 2016 but granted in 2025, the plaintiff asserts that widely deployed models, including Anthropic’s Claude and OpenAI’s ChatGPT, infringe on protected neural network mechanisms. For companies utilizing similar attention mechanisms in their proprietary models, this creates a direct liability risk where past development choices may now trigger retroactive damages claims. The specific case numbers, 1:26-cv-01454 for Anthropic and 1:26-cv-01455 for OpenAI, signal a formalized legal battle that could set precedents for how legacy patents are applied to modern AI implementations.

Operational risks extend beyond immediate litigation costs to potential disruptions in product availability and licensing negotiations. If courts validate the claims that these legacy patents cover fundamental attention mechanisms, developers may face injunctions or mandatory licensing fees that alter the economics of model deployment. The fact that the patents were granted recently despite early filing dates suggests a "submarine patent" strategy, where the threat emerges only after the technology has become industry-standard. This timing forces companies to assess whether their current infrastructure is vulnerable to similar claims, potentially requiring costly redesigns or defensive patent acquisitions to mitigate exposure.

Key risks for AI developers include:

  • Retroactive Damages: Exposure to compensation claims for past use of attention mechanisms that were not previously considered infringing.
  • Injunction Threats: Potential court orders restricting the sale or deployment of models that rely on the disputed neural network architectures.
  • Licensing Costs: The possibility of being forced to negotiate licenses with patent holders like Flynn Artificial Intelligence Technologies, increasing operational expenses.
  • Due Diligence Gaps: The need for enhanced IP audits to identify other legacy patents that may have been granted recently and cover core AI components.

Defensive Strategies for In-House Counsel

Immediate Defensive Playbook for AI Developers

For in-house counsel at Anthropic and OpenAI, the immediate priority following the Delaware federal court filings by Flynn Artificial Technologies LLC is to initiate a rigorous prior art search focused on the specific attention-based neural network claims. Given that the patents in question were filed in 2015 and 2016 but only granted in 2025, legal teams must aggressively investigate whether the underlying technology was publicly disclosed or commercially used prior to those filing dates. This involves scrutinizing academic publications, open-source repositories, and industry white papers from the mid-2010s to determine if the claims lack novelty or non-obviousness, which could form the basis for invalidity defenses or inter partes review petitions at the USPTO.

Concurrently, a targeted freedom-to-operate (FTO) analysis is essential to map the specific technical implementations of Claude and ChatGPT against the five asserted patents. Counsel must determine whether the accused products utilize the exact mechanisms described in the claims or if they rely on alternative architectures that do not infringe. This technical-legal mapping will clarify the strength of the infringement allegations and help quantify the potential liability exposure, allowing the company to decide whether to pursue early settlement discussions or prepare for a protracted litigation strategy.

Key immediate actions for legal teams include:

  • Prior Art Deep Dive: Commission expert searches for pre-2015/2016 disclosures related to attention mechanisms to challenge patent validity.
  • FTO Technical Mapping: Engage engineering leads to document how current model architectures differ from the specific claims in the Flynn patents.
  • Litigation Hold: Implement strict document preservation protocols for all engineering, product, and executive communications related to the development of the accused models.
  • Licensing Assessment: Evaluate the feasibility of negotiating a license with Flynn Artificial Technologies to mitigate business disruption, while simultaneously preparing invalidity arguments as leverage.

Long-Term Implications for AI IP Portfolios

Strategic Shifts in Patent Acquisition and Portfolio Defense

The litigation brought by Flynn Artificial Technologies LLC against Anthropic and OpenAI highlights a critical vulnerability in the current AI IP landscape: the threat posed by legacy patents filed years before the technology became commercially viable. Since the patents at issue were filed in 2015 and 2016 but only granted in 2025, they represent a "sleeping giant" risk that many AI developers may have overlooked during their initial freedom-to-operate analyses. This timeline suggests that future patent filing strategies must extend beyond immediate product development cycles to account for long-pending applications that could emerge as enforcement tools years later. Companies may need to adopt more aggressive pre-grant monitoring systems to identify and challenge potentially disruptive claims before they mature into enforceable assets.

This case is likely to accelerate the trend of defensive patent aggregation within the AI sector. As the risk of litigation from non-practicing entities holding broad, foundational neural network patents increases, major AI firms may seek to pool resources or acquire portfolios that cover core architectural elements like attention mechanisms. Rather than relying solely on internal R&D patents, which often focus on specific implementations, companies may look to defensive aggregators or collaborative licensing groups to create a shield against broad claims. This shift could lead to a more consolidated IP market where access to foundational patents is negotiated collectively, reducing the individual liability exposure for smaller AI startups who lack the resources to litigate against well-capitalized patent holders.

Key implications for AI IP strategy include:

  • Extended Monitoring Horizons: Legal teams must track patent applications filed years ago, particularly those related to foundational architectures, as they may surface as litigation threats long after the technology is deployed.
  • Defensive Aggregation: Increased reliance on patent pools or defensive aggregators to mitigate the risk of infringement suits from entities holding broad, legacy patents.
  • Proactive Invalidity Searches: Conducting deeper prior art searches on older filings to identify vulnerabilities in patents that were granted recently but based on earlier applications.
  • Licensing Negotiations: A potential rise in cross-licensing deals focused on core neural network components to avoid costly litigation over fundamental technologies.

Frequently Asked Questions

Who is suing Anthropic and OpenAI for patent infringement?

Flynn Artificial Intelligence Technologies LLC, a Texas-based patent holding company, filed the lawsuits in the U.S. District Court for the District of Delaware on November 12, 2026. The company alleges that Anthropic's Claude models and OpenAI's ChatGPT infringe on five neural network patents.

What specific patents are involved in the lawsuit against AI companies?

The lawsuit targets five patents related to attention-based neural networks, which were co-invented by Stephen Flynn and Naomi Moneyfeeny. These patents were originally filed in 2015 and 2016 before being granted in 2025.

What are the case numbers for the Anthropic and OpenAI patent lawsuits?

The case against Anthropic is filed under case number 1:26-cv-01454, while the case against OpenAI is filed under 1:26-cv-01455. Both cases are currently pending in the Delaware federal court, where the plaintiff is seeking damages for alleged infringement.

Sources

Adopt AI in legal work, carefully

MeshLaw is an AI case-management tool for lawyers. No hallucinations, fully verifiable.

Explore MeshLaw →

← Back to all briefings

AI case management for lawyers — MeshLaw Try it free →