Autodesk AI Transparency Cards

The top of each AI transparency card identifies the Autodesk product that includes the AI feature, the name of the AI feature, the release status when applicable, and what this feature does. 

Release status may include: 

• Beta: A pre-release version of the feature that may have limited functionality and is subject to change. 
• Tech Preview: An early-access version of the feature that is still in development and is subject to change or removal. 

“What this AI does” summarizes what the AI feature is designed to do when used in the product.

The Data handling section provides information about what data is handled by the AI feature and how it is handled. This includes: 

• Training data: Refers to data used to train the AI model behind the feature. 
• Customer choice: Indicates whether customers have a choice about using Autodesk AI features.

Data types show the categories of data that Autodesk used to train the AI model powering the feature. The listed types reflect common, non-proprietary, recurring patterns within each data type group rather than customer-specific data.  

• Geometry: geometric shapes, such as lines, dots, and figures. 
• Drawings: technical or design drawings, such as plans, diagrams, sketches, or CAD drawings. 
• Documents: documents, such as specifications, reports, forms, schedules, or other project or product documentation. 
• Image: static visual data, such as photos, renderings, screenshots, scans, or image-based references. 
• Text: text-based information, such as descriptions, labels, notes, comments, or extracted text. 
• Prompts: user-entered instructions or requests that guide an AI feature’s output. 
• Video: video, animation sequences, or screen recordings. 
• Audio: sound-based data, such as speech, voice, music, sound effects, or other audio recordings. 
• Environmental data: information about environmental or site conditions, such as location context, terrain, weather, lighting, energy, carbon, or other environmental factors. 
• Financial data: cost, budget, pricing, estimate, or other finance-related information. 
• Motion data: movement or trajectory information, such as motion capture data, character rigging, robotic/machine toolpaths, animation rigs, and simulated or sensor-tracked motion. 
• N/A (third-party model): The feature uses a third-party model trained independently by the provider.  
• N/A (no model training): The technique used by the feature does not rely on a model that requires a training dataset.

Data sources indicate the source of data used to train the AI model powering the feature. They include: 

• Customer content: Data that a customer or their authorized users create or import into Autodesk products and services. 
• Licensed third-party: Data acquired from external organizations under agreements that permit the relevant use for this feature, subject to applicable license restrictions. 
• Open source: Data made available under an open license and used as allowed by that license. 
• Synthetic data: Data artificially generated to support model training. 
• N/A (third-party model): The feature uses a third-party model trained independently by the provider.  
• N/A (no model training): The technique used by the feature does not rely on a model that requires a training dataset.

Personal data indicates whether personal data is present in the training datasets used to train the AI model powering the feature. 

• Used: Personal data is present in the training datasets used to train the AI model. 
• Not used: Based on Autodesk’s review, personal data isn’t expected to be present in the training datasets used to train the AI model. 
• N/A (third-party model): The feature uses a third-party model trained independently by the provider. 
• N/A (no model training): The technique used by the feature does not rely on a model that requires a training dataset.

Feature classification indicates the category of AI feature to help customers understand whether the feature is designed for everyday productivity or more complex tasks.

• Standard: Designed to support everyday productivity and workflow efficiency by optimizing, automating, or accelerating day-to-day tasks.  
• Advanced: Designed for more complex AI-assisted workflows that may generate or explore options for customer designs, plans, selections, or content. 

Choice format indicates whether and how customers can control the use of the AI feature.

• Turn off: Customers can turn the feature off at supported levels. 
• Always on: The feature cannot be turned off.

Human oversight indicates whether users can review, update, or approve generated outputs before a final action is taken.

• Yes: Users can review, edit, or approve generated outputs before a final action is taken.
• No: Generated outputs are applied automatically without a required human review step.

Even when human oversight is available, users should review AI outputs for accuracy, completeness, and suitability for their intended use.

The Known limitations section indicates constraints or considerations customers should be aware of when using the AI feature.

• Not for safety-critical use: The feature and its outputs should not be relied on to fulfill a safety function. 
• Some actions are irreversible: Some actions taken by this feature cannot be undone. Users should review outputs carefully before applying them. 
• Results may vary based on inputs: Output quality and accuracy may depend on the quality, completeness, format, or specificity of the inputs provided. 
• May produce inconsistent results: The feature may generate outputs that vary across similar or repeated uses.

Built-in safeguards are security and data protection measures designed for a specific AI feature to help reduce risks associated with that feature. These safeguards operate alongside the product’s general security measures and are not intended to describe or replace those measures.

• Encryption: Autodesk uses encryption for data in transit and at rest where supported and as described in applicable security documentation. 
• Tokenization: Certain information is replaced with tokens to represent the original data without exposing it during processing. 
• Anonymization: Data is irreversibly altered so it can no longer be linked back to an individual.  
• De-identification: Identifying details are removed or obscured, but the data may be re-linked to an individual under certain conditions. 
• On-device processing: Data is processed locally on the user's device instead of being processed on external servers.

Model provider identifies who develops or provides the underlying AI model that powers the feature.

• Autodesk: The AI model powering this feature is developed and maintained by Autodesk.
• Third-party: The AI model powering this feature is provided by an external organization under agreement with Autodesk. 
• Autodesk, third-party: The feature uses a combination of Autodesk-developed models and models provided by external organizations under agreement with Autodesk.

Primary technique identifies the main technical approach used to develop or power the AI feature. 

• Autoencoders: Compress data into a simpler representation and reconstruct it to identify important patterns or detect unusual results.
• Classification: Categorizes items based on patterns learned from previous examples.
• Computer Vision: Analyzes visual data such as images, video, or drawings to identify objects, shapes, patterns, or features.
• Convolutional Neural Network: Processes visual data and detects patterns such as edges, textures, and shapes in images.
• Encoding: Converts data into a format that machine learning models can process so the information can be searched, compared, grouped, or used by another model.
• Generative Design: Generates potential design options based on customer-defined goals, parameters, and constraints.
• Evolutionary Algorithm: Improves possible solutions multiple iterations based on defined criteria.
• Graph Neural Network: Analyzes connections between data points, such as components or dependencies within a design, to understand relationships, identify patterns, or support predictions.
• Predictor: Uses historical patterns  to estimate or predict future values or outcomes.
• Recommendations: Suggests relevant options, actions, or content based on available information or user input.
• Retrieval-Augmented Generation (RAG): Retrieves relevant information and uses it to improve generated responses.
• Transformer: Processes complex relationships in data to support tasks such as understanding input, answering questions, translation, and text generation.
• Diffusion Transformer (DiT): Creates new data by progressively refining an initial, unstructured pattern into a structured output.