AI & Generated Content Disclaimer

This AI and Generated Content Disclaimer ("Disclaimer") applies to all outputs produced by the RadMah AI platform, including but not limited to: — Synthetic tabular data generated by any RadMah AI engine — ICS security simulation and SCADA telemetry data — Protocol and network traffic data — Evidence Bundles and their constituent artifacts — Outputs produced by the AI Orchestrator feature — Outputs produced by the Agentic AI Data Scientist feature — Any other content, data, analysis, recommendation, or artefact generated by or with assistance from AI or algorithmic systems within the Service This Disclaimer supplements the Terms of Service (www.radmah.ai/legal/terms) and the Acceptable Use Policy (www.radmah.ai/legal/aup). In the event of any conflict, the Terms of Service govern.

2.1 Synthetic Nature. All data produced by RadMah AI's generation engines is entirely synthetic. It does not represent or describe real individuals, real events, real organisations, real infrastructure, or real operational states, unless explicitly configured to replicate statistical properties of a Reference Dataset (and even then, the output is statistically similar, not reproduced). 2.2 AI Model Characteristics. The AI Orchestrator and Agentic AI Data Scientist features incorporate large language models (LLMs) operated by third-party AI providers (including but not limited to OpenAI, Anthropic, and Google). These models have the following inherent characteristics that you must understand before relying on their outputs: (a) Hallucination. LLMs can generate content that is plausible-sounding but factually incorrect, fabricated, or inconsistent with real-world knowledge. This includes hallucinated references to standards, regulations, academic papers, data structures, or technical specifications. (b) Non-Determinism. Unlike RadMah AI's deterministic data generation engines (which produce bit-for-bit identical output for the same Contract K and seed), LLM-based features produce non-deterministic outputs. The same prompt submitted twice may produce materially different results. (c) Knowledge Cutoffs. LLMs have training data cutoffs and do not have access to real-time information, current regulatory guidance, or up-to-date industry standards unless explicitly provided as context. (d) Bias. LLMs may reflect biases present in their training data. Outputs may not be balanced, representative, or appropriate for all use cases. (e) Contextual Errors. LLMs may misinterpret ambiguous instructions, make incorrect assumptions about context, or produce outputs that are technically or legally inappropriate for your specific situation. 2.3 Deterministic Engine Outputs. Data generated by RadMah AI's deterministic engines (Mock Data, Synthesize, Virtual SCADA Simulator, ICS Security Simulator, Constrained Synthesis) is algorithmically produced and is not subject to hallucination. However, these outputs are statistical constructs and do not represent ground truth. Fitness for any specific technical, regulatory, or commercial purpose must be independently assessed by qualified professionals.

3.1 RadMah AI makes no representation or warranty, express or implied, that any output from the Service (including Generated Data, Evidence Bundles, AI Orchestrator outputs, or Agentic AI Data Scientist outputs) is: (a) Accurate, complete, up-to-date, or free from errors; (b) Fit for any particular purpose, including regulatory submission, legal proceedings, financial modelling, clinical research, or safety-critical system design; (c) Compliant with any specific technical standard, regulatory requirement, or legal obligation; (d) Representative of real-world phenomena, systems, or behaviours; (e) Free from bias, hallucination, or systematic error. 3.2 Generated synthetic data should be treated as a starting point for analysis, experimentation, or augmentation — not as a substitute for authoritative, real-world data or expert professional advice. 3.3 The presence of a BLAKE3 Cryptographic Seal in an Evidence Bundle establishes the cryptographic integrity and reproducibility of the data generation job — it does not constitute a warranty of the accuracy, completeness, or regulatory suitability of the Generated Data itself.

You are solely responsible for: 4.1 Verification. Independently verifying the accuracy, completeness, and suitability of any Generated Data or AI output before using it in any consequential context. This obligation applies regardless of the plans, tiers, or features used, and regardless of any Evidence Bundle produced. 4.2 Expert Review. Engaging qualified domain experts — including but not limited to data scientists, cybersecurity professionals, legal counsel, regulatory affairs specialists, actuaries, or medical professionals — to review AI-generated outputs before relying on them for professional, commercial, regulatory, or safety-critical purposes. 4.3 Regulatory Fitness. Determining independently whether Generated Data or Evidence Bundles meet the requirements of any specific regulatory body, standard, or legal framework in your jurisdiction. RadMah AI does not provide regulatory, legal, medical, or financial advice. 4.4 Model Validation. If you use Generated Data to train, fine-tune, or validate AI/ML models, you are responsible for independently validating those models before deploying them in production or consequential contexts. 4.5 No Reliance in Safety-Critical Contexts. You must not rely solely on Generated Data or AI-generated outputs in any safety-critical context — including real-time industrial control, clinical decision-making, autonomous vehicle operation, public safety systems, or defence applications — without independent verification, expert oversight, and appropriate validation processes.

5.1 Recommendation Risk. The AI Orchestrator and Agentic AI Data Scientist features may make recommendations regarding schema design, engine selection, parameter configuration, data quality, or statistical methodology. These recommendations are generated by LLMs and may be incorrect, incomplete, or unsuitable for your specific use case. All recommendations must be reviewed and validated by qualified personnel before implementation. 5.2 Prompt Injection Risk. AI features that process user-provided inputs (including Reference Dataset column names, schema descriptions, or freeform instructions) may be susceptible to prompt injection attacks. RadMah AI implements technical mitigations but cannot guarantee immunity. You should not include sensitive, confidential, or privileged information in AI prompts without understanding this risk. 5.3 No Professional Advice. Nothing generated by the AI Orchestrator or Agentic AI Data Scientist constitutes legal, financial, medical, regulatory, cybersecurity, or other professional advice. All outputs are informational starting points only. 5.4 Third-Party AI Providers. When using AI features: — If you use RadMah AI's platform-managed API key, your prompts are processed by the relevant third-party AI provider (OpenAI, Anthropic, or Google) under RadMah AI's enterprise agreement with that provider. Prompts are not used by the AI provider to train their models under RadMah AI's agreements. — If you use a bring-your-own (BYO) API key, your prompts and outputs are governed entirely by your own agreement with the AI provider. RadMah AI is not responsible for any processing performed by the AI provider under your BYO key. 5.5 Output Variability. AI-generated outputs are inherently variable. Repeating the same workflow may produce different results at different times. Do not rely on AI feature outputs as a source of deterministic, reproducible data — use the deterministic generation engines for that purpose.

6.1 Evidence Bundles are designed to provide cryptographic proof of data provenance, job configuration, determinism, and integrity — not to certify regulatory compliance, statistical validity, or fitness for any specific regulatory submission. 6.2 RadMah AI has not received, and does not represent that it has received, approval from any regulatory authority (including the FDA, EMA, FCA, SEC, CFTC, NRC, NERC, or any other regulatory body) for the use of its Generated Data or Evidence Bundles in formal regulatory submissions. 6.3 Customers who intend to use Generated Data or Evidence Bundles in formal regulatory submissions are responsible for: (a) independently determining whether the regulatory authority will accept synthetic data; (b) engaging regulatory affairs specialists; (c) conducting all required validation and statistical testing; and (d) ensuring compliance with all applicable regulatory guidance on the use of synthetic data. 6.4 The RadMah AI Privacy Report (artifact #6 in the Evidence Bundle) provides a technical privacy risk assessment generated by algorithmic means. It does not constitute a Data Protection Impact Assessment (DPIA) as required under GDPR Article 35, nor does it substitute for legal or privacy counsel's review.

RadMah AI's platform may constitute an "AI system" within the meaning of Regulation (EU) 2024/1689 (the EU AI Act). 7.1 Risk Classification. RadMah AI's synthetic data generation engines are not currently classified as high-risk AI systems under Annex III of the EU AI Act. RadMah AI monitors regulatory developments and updates this disclosure as the regulatory framework evolves and as additional guidance from the European AI Office is issued. 7.2 Transparency. RadMah AI provides technical documentation of its generation algorithms (available to Enterprise customers under NDA) that may support customers' own AI Act compliance obligations. 7.3 Prohibited AI Practices. RadMah AI's platform is not designed or intended for any AI practice prohibited under EU AI Act Article 5, including subliminal manipulation, social scoring by public authorities, or real-time biometric identification in public spaces. Use of RadMah AI's platform for any prohibited AI practice is a violation of the Acceptable Use Policy. 7.4 Customer Obligations. Customers who deploy AI systems trained on RadMah AI-generated data and who are themselves subject to EU AI Act obligations are solely responsible for their own compliance with those obligations.

8.1 Ownership. You own the Generated Data and Evidence Bundles produced by your Account. RadMah AI makes no claim to ownership of your Generated Data. 8.2 Third-Party IP Risk. While RadMah AI's generation engines produce synthetic data algorithmically, RadMah AI cannot guarantee that Generated Data will not bear a superficial resemblance to, or inadvertently replicate elements of, proprietary third-party datasets or copyrighted works. You are solely responsible for conducting any necessary intellectual property clearance before using Generated Data in commercial or published contexts. 8.3 No Indemnity for IP Claims. RadMah AI's indemnification obligations under the Terms of Service do not extend to claims arising from third parties alleging that Generated Data infringes their intellectual property rights. Such claims are the Controller's responsibility under the indemnification provisions of the Terms of Service.

To the maximum extent permitted by applicable law, RadMah AI's liability for any loss, damage, claim, or expense arising from: — Errors, inaccuracies, or hallucinations in Generated Data or AI feature outputs; — Your reliance on Generated Data for any professional, regulatory, commercial, or safety-critical purpose; — The use of Generated Data to train AI models that subsequently cause harm; — Any failure of an Evidence Bundle to be accepted by a regulatory body, court, or third party; — Any intellectual property claim relating to Generated Data; is subject to the aggregate liability cap and consequential damages exclusion set out in Section 17 (Limitation of Liability) of the Terms of Service. This limitation of liability reflects a fundamental allocation of risk. RadMah AI generates data at scale for many purposes it cannot foresee or control. You, as the party deploying that data in a specific context, are best placed to assess fitness for purpose and must take responsibility for that assessment.

If you have questions about the limitations of AI-generated data for your specific use case, our technical team is available to discuss: Technical queries: support@radmah.ai Privacy and regulatory queries: privacy@radmah.ai Legal queries: legal@radmah.ai Enterprise consultation: contact@radmah.ai