Iron Team Clarity, Inc. DBA Iron Gorilla California Generative AI Training Data Transparency Disclosure
Disclosure version 1.0
| Scope and legal basis. This disclosure is intended to address California Civil Code sections 3110 and 3111 (AB 2013). It describes data used directly by Team Clarity, Inc. to develop, test, validate, fine-tune, or distill the Iron Gorilla generative AI systems and services identified below. It does not present an upstream model developer’s training data as Iron Gorilla’s own. |
1. System Identification
| Field | Disclosure |
|---|---|
| System or service name | Iron Gorilla Generative Model Services, including: (i) AI Agent Builder; (ii) AI Policy Builder; (iii) the Homepage AI Business Consultant; (iv) the Iron Gorilla General Language Model beta; and (v) Managed Voice Agents at the system-orchestration and validation layer. Customer-configured agents may also use third-party inference providers. |
| Version / release | Current production and beta releases available as of July 29, 2026. Iron Gorilla uses continuously versioned model and service releases. Model-specific identifiers and release manifests are maintained internally, and a material retraining, fine-tune, distillation, or functionality change triggers review and an updated disclosure. |
| Developer legal entity | Team Clarity, Inc., doing business as Iron Gorilla. Team Clarity designs, codes, produces, and substantially modifies the Iron Gorilla-developed systems and services identified above and is the developer for those systems within the scope of California Civil Code section 3110. Team Clarity is not the developer of third-party upstream models merely integrated into the platform. |
| Public availability date | The covered Iron Gorilla systems and services were released after January 1, 2022 and were available to California users by July 29, 2026. Exact feature-level launch dates are maintained in Iron Gorilla release records and should be included in any version-specific archived disclosure. |
| Modalities | Text and code for Agent Builder, Policy Builder, the Homepage AI Business Consultant, and the Iron Gorilla language-model beta; synthetic audio for Managed Voice Agents. Image and broader multimodal capabilities remain under research and development unless separately identified as publicly available. |
| Material modification or fine-tune | Yes. Iron Gorilla develops task-specific coding and language models through distillation and further training of appropriately licensed open-weight foundation models, together with Iron Gorilla-developed training methods, datasets, evaluations, and deployment controls. Iron Gorilla also performs system-level testing, validation, orchestration, governance, and integration. Where a service uses an upstream model without Iron Gorilla retraining that model, the upstream model is identified separately below. |
2. Dataset Summary
| Required topic | Product-verified disclosure |
|---|---|
| Sources or owners | The development data used directly by Iron Gorilla may include: appropriately licensed open-weight model artifacts and associated documentation from their respective owners; Iron Gorilla-owned or proprietary agent instructions, policy logic, workflows, tool schemas, code, test prompts, expected outputs, evaluation rubrics, and human-reviewed examples; public-domain or permissively licensed technical materials and code where their terms allow the relevant use; and synthetic examples generated for distillation, task coverage, safety, and evaluation. Release-specific source categories are recorded in internal dataset and model manifests. |
| How the datasets further the intended purpose | The datasets are selected to improve generation of agent instructions, executable workflows, governance policies, code, tool-use plans, conversational responses, and safe operating behavior. Evaluation and red-team datasets are used to measure correctness, policy validity, tool selection, refusal behavior, safety, reliability, and robustness before and after release. |
| Number of data points or approximate scale | Iron Gorilla measures directly assembled post-training data in examples and tokens, with counts varying by model and release. Exact release-specific counts are maintained in internal dataset manifests. |
| Data types and labels | Primarily text and code, including natural-language requirements, agent instructions, policy rules, structured workflows, tool and API schemas, code snippets, instruction-response pairs, synthetic dialogues, expected outputs, test cases, and metadata. Labeled data may include quality, correctness, safety, policy-validity, tool-selection, approval-routing, refusal, and human-preference or acceptance labels. Unlabeled data may include general technical text, code, model outputs, and structured examples. |
| Copyrighted, trademarked, or patented material | Yes, some source materials or upstream model artifacts may be protected by copyright or may contain trademarks. Iron Gorilla uses such materials only under applicable open-weight, open-source, commercial, or other licenses and permissions, or another lawful basis identified in the relevant release manifest. Datasets are not represented as entirely public domain. Iron Gorilla does not intentionally curate patented inventions as a training-data category, although software and model-development technologies may be subject to third-party patent rights. |
| Licensed or purchased data | Yes. Iron Gorilla uses appropriately licensed open-weight foundation models and may use permissively licensed code, documentation, datasets, or commercially licensed services. The applicable license category and source are recorded for each model release. Iron Gorilla does not claim ownership of upstream model training data and does not copy an upstream developer’s disclosure as its own. |
| Personal information | Iron Gorilla does not intentionally collect or curate personal information, sensitive personal information, or protected-category data for model training. Public, licensed, or upstream materials may incidentally contain personal information or aggregate consumer information. Where Iron Gorilla directly curates data, it applies source controls, filtering, redaction, and exclusion procedures designed to reduce unnecessary personal information, credentials, secrets, and other sensitive content. |
| Cleaning, filtering, and modification | Depending on the release, processing may include deduplication; normalization and formatting; tokenizer and schema conversion; source and license review; removal or masking of personal information, credentials, secrets, malware, unsafe content, malformed code, and low-quality or duplicate examples; quality scoring; human review; safety filtering; balancing and augmentation; train/evaluation separation; and transformation of teacher-model outputs into task-specific distillation or instruction-tuning examples. |
| Collection period | The directly assembled Iron Gorilla datasets were collected and developed from the commencement of the applicable internal model-development program through July 29, 2026, and curation is ongoing. Source-level date ranges vary by dataset and are maintained in internal dataset manifests. |
| First use in training | The datasets were first used during the internal development and beta-validation period preceding the 2026 public and customer beta releases. |
| Synthetic data | Yes. Iron Gorilla may use synthetic instruction-response, code, policy, workflow, edge-case, and evaluation examples generated under Iron Gorilla’s control using appropriately licensed models or Iron Gorilla-developed systems. Synthetic data supports distillation, task coverage, safety testing, augmentation of sparse examples, and evaluation. The proportion varies by model release and is recorded internally where known. |
| User data or customer data | Customer prompts, outputs, logs, files, retrieved content, and other customer data are not used to train Iron Gorilla-developed models by default. Such data may be used for service delivery, security, support, troubleshooting, abuse prevention, and performance monitoring. Model training using customer data requires the customer’s explicit written authorization or opt-in, an applicable contractual and privacy basis, and release-specific documentation. |
3. Exclusions and Upstream Models
Security-only exclusion. Iron Gorilla does not rely on the California Civil Code section 3111 security-and-integrity exclusion for Agent Builder, Policy Builder, the Homepage AI Business Consultant, or the Iron Gorilla language-model beta because those offerings generate synthetic content and are not limited to a sole security-and-integrity purpose.
Upstream inference providers. Customers may configure agents to use third-party inference providers, including OpenAI, Anthropic, and xAI. Managed Voice Agents currently use OpenAI’s GPT-Realtime-2.1 as an upstream speech-to-speech model. Team Clarity provides the branded system, orchestration, interfaces, tool integrations, governance, runtime, and deployment mechanisms, but it does not represent that it developed or trained those upstream models. Training-data information for an upstream model should be obtained from the upstream developer’s own disclosure and model documentation.
OpenAI upstream disclosure: Training Data Summary Pursuant to California Civil Code Section 3111
Iron Gorilla-developed models. Agent Builder and Policy Builder use an internally developed Iron Gorilla coding model. The Homepage AI Business Consultant and selected beta use cases may use an internally developed Iron Gorilla language model. Those models were developed through distillation and further training of appropriately licensed open-weight foundation models, together with Iron Gorilla-developed data, methods, evaluations, and controls. The disclosure above covers data used directly by Iron Gorilla in that development and does not attribute the upstream model developers’ original training datasets to Iron Gorilla.
System-level testing and validation. Because California’s definition of training includes testing, validation, and fine-tuning, Iron Gorilla treats system-level evaluation datasets used for its own branded services as within this disclosure even when an underlying model is supplied by a third party.
4. Updates
Iron Gorilla reviews and updates this disclosure before making a new covered generative AI system or service, or a substantial modification, publicly available to Californians. Update triggers include a new model or service release; material retraining, distillation, or fine-tuning; a material change in functionality or performance; addition of a new data source or dataset category; a change in customer-data training practices; or a material change in cleaning, filtering, licensing, privacy, or synthetic-data practices.
For each update, Iron Gorilla will:
review the applicable model and dataset manifests;
obtain Product, Engineering, Privacy, Security, and Legal approval as appropriate;
publish the updated disclosure before the covered release is made available to Californians;
assign a new disclosure version and effective date; and
retain an archived copy of the prior disclosure and the supporting internal approval record.
Version history
| Version | Date | Summary |
|---|---|---|
| 1.0 | July 29, 2026 | Initial consolidated disclosure for Iron Gorilla-developed generative model services and system-level use of upstream models. |
5. Contact
| Questions | legal@teamclarity.ai |
|---|---|
| Product owner | Jacob Hartmann, Chief Executive Officer, Iron Gorilla |
| Management approval date | July 29, 2026 |