Orca Security Ltd v. Wiz Inc

Prosecution History analysis

Examiner Search analysis

Invalidity Framework analysis

Petition

Petition Analysis

## 1. Case Identification
- Case #: [IPR2025-01087](https://ai-lab.exparte.com/case/ptab/IPR2025-01087)
- Patent #: [12,003,529](https://ai-lab.exparte.com/docs/ptab/IPR2025-01087/1001)
- Filed: June 4, 2025
- Petitioner(s): [Orca Security](https://ai-lab.exparte.com/party/orca-security-ltd) Ltd.
- Patent Owner(s): [Wiz](https://ai-lab.exparte.com/party/wiz-inc), Inc.
- Challenged Claims: 1-17

## 2. Patent Overview
- Title: Detecting Cybersecurity Risks in Artificial Intelligence Models
- Brief Description: The [’529](https://ai-lab.exparte.com/docs/ptab/IPR2025-01087/1002) patent discloses a method and system for identifying cybersecurity risks associated with artificial intelligence (AI) models operating within a cloud computing environment. The technology involves creating an inspectable copy (e.g., a snapshot) of a cloud resource, inspecting that copy to locate AI models, generating a graph-based representation of the environment and the identified models in a security database, analyzing the models for risks, and initiating mitigation actions based on the findings.

## 3. Grounds for Unpatentability

#### Ground 1: Obviousness over Shua and Lang - Claims 1-17 are obvious over Shua in view of Lang.
  - **Prior Art Relied Upon:** Shua (Patent [11,489,863](https://ai-lab.exparte.com/docs/ptab/IPR2025-01087/1004)) and Lang (Application # [2024/02022405A1](https://ai-lab.exparte.com/docs/ptab/IPR2025-01087/1006)).
  - **Core Argument for this Ground:**
    - **Prior Art Mapping:** Petitioner argued that the combination of Shua and Lang rendered all challenged claims obvious under 35 U.S.C. §103. Shua was presented as teaching a comprehensive cloud security system that performs agentless scanning by taking snapshots of cloud resources (the claimed "inspectable disk") to create a complete inventory. Shua further taught generating a graph-database representation of the entire cloud environment, mapping assets and their interconnections, and inspecting them for general cybersecurity risks like known vulnerabilities (CVEs) and misconfigurations. However, Petitioner asserted that Shua did not explicitly teach methods for specifically identifying and analyzing AI models.

      Lang allegedly supplied these missing elements. Petitioner argued Lang taught methods for analyzing computing systems, including virtual machine disks, to specifically detect, analyze, and harden AI systems. Lang disclosed inspecting for AI components by, for example, identifying library imports (e.g., PyTorch, TensorFlow) on a disk, which corresponds to the claimed "inspecting the inspectable disk for an AI model." Lang also taught analyzing these models for AI-specific vulnerabilities, such as data poisoning and adversarial attacks, and creating abstract representations of the AI models and their associated risks.

      By combining the references, a POSITA would arrive at the claimed invention: using Shua's foundational system for snapshotting and mapping a cloud environment, and integrating Lang's specific teachings to inspect those snapshots for AI models, represent them within Shua's security graph, analyze them for AI-specific risks, connect those risks to the AI model nodes in the graph, and initiate mitigation actions as taught by both references. The arguments for apparatus claims 9 (non-transitory medium) and 10 (system) and their dependent claims were asserted to be unpatentable for the same reasons as their corresponding method claim limitations.

    - **Motivation to Combine:** Petitioner asserted that a Person of Ordinary Skill in the Art (POSITA) would have been motivated to combine the teachings of Shua and Lang. The primary motivation was to improve Shua's system to be more comprehensive. Shua provided an express goal of scanning "100% of an organization's assets," and Lang taught that AI models are a type of asset increasingly prevalent and particularly prone to unique security risks. A POSITA would therefore combine Lang's known technique for AI model analysis with Shua's similar, broader system to achieve a more complete and effective security posture, which was a recognized goal in the field.

    - **Expectation of Success:** A POSITA would have had a reasonable expectation of success in making the combination. Both references operate in the same field of cloud and system cybersecurity and address analogous problems. Integrating Lang's AI-specific inspection modules into Shua's broader scanning and data-modeling framework was argued to be a predictable implementation involving, at most, minor software modifications.

    - **Key Aspects:** The core of the challenge relied on the argument that the [’529](https://ai-lab.exparte.com/docs/ptab/IPR2025-01087/1002) patent’s contribution—adding AI model awareness to a general cloud security platform—was not inventive. Petitioner contended this was merely the application of a known technique (Lang's AI analysis) to a known system (Shua's cloud scanner) to yield predictable results. The amendments made during prosecution to secure allowance (adding the "inspectable disk" and "inspecting for an AI model" steps) were argued to be features already well-known from Shua and the combination of Shua with Lang, respectively.

## 4. Relief Requested
- Petitioner requests institution of an inter partes review (IPR) and cancellation of claims 1-17 of Patent [12,003,529](https://ai-lab.exparte.com/docs/ptab/IPR2025-01087/1001) as unpatentable.

Orca Security targets Wiz's AI model security patent: Shua & Lang prior art cited against all 17 claims as obvious over agentless cloud scanning.