Who Owns the Risk When Generative AI Assists with Plant Data Models and Diagrams?

Who owns the risk?

When generative AI assists with plant data models, calculations, or diagrams

Generative AI-based design and engineering tools can provide suggestions to complete data models, calculate missing values, and partly even assist in generation of engineering diagrams. From a legal perspective, however, the outcomes are not yet differentiated from what humans produce. This article collects what is available as of now on this topic and lays out a plausible roadmap for companies dealing with AI-generated data and diagrams.

1. Contract law — your company is still on the hook

1.1 Privity rules remain unchanged

Whether you sign an EPC contract in Texas or a turnkey agreement in Bavaria, the counter-party will sue your firm if major mistakes in documentation cause damages—even if those drawings were auto-generated by an LLM-powered CAD plug-in. AI is legally just another subcontractor’s tool.

1.2 Modern AI clauses shift — but do not erase — risk

“As-is” plus zero-warranty language from AI vendors is now common in many MSAs. Enterprises often pay a premium for a “no data retention” or “protected IP” tier, but the fine print still typically limits the provider’s liability to low six figures or the last 12 months’ fees.

Indemnities flow both ways. Large-model providers (Microsoft, Google, OpenAI) now offer IP-infringement indemnities—yet require strict compliance with usage guidelines and may claw back cover if you tweak system prompts. Conversely, smaller SaaS vendors often make the customer indemnify them for “hallucination” losses.

AI-related amendments are becoming routine in contracts. The 2024 NVCA Stock Purchase Agreement, for instance, forces a target company to warrant that any internal AI use does not contaminate trade secrets and complies with privacy laws.

1.3 European “fair-terms” scrutiny

EU contracts cannot exclude liability for gross negligence or public-safety failures. Clauses that purport to give an AI user zero recourse could be struck down under the Unfair Contract Terms Directive—especially in B2C contexts.

Summary: You can draft AI-specific warranties and clauses, but you should assume that, in court, your firm—not the model vendor—will most probably be first in line.

2. Professional liability — engineers cannot delegate the duty of care

2.1 U.S. “responsible charge”

Every state board requires a licensed Professional Engineer to review, modify where necessary, and seal safety-critical plans. NCSEA’s 2024 town-hall makes the comparison explicit: treat AI output the way you treat a junior intern’s sketches—verify and document. Signing off without review is negligence per se.

2.2 ASCE Policy Statement 573

Adopted July 18, 2024, PS 573 states bluntly that “AI cannot replace the professional judgement of a PE” and that the engineer remains “accountable for project planning, design and public welfare.” Many insurers have already added AI-usage questionnaires to renewal applications.

2.3 EU national codes echo the rule

In Germany the Bauordnungen require a Nachweisberechtigter (qualified engineer) to validate structural calculations. France’s Code de la Construction imposes similar duties. Failing to insert a human validation step could breach statutory “obligation de résultat” for safety.

Summary: “Human-in-the-loop” is not optional for licensure; log every review decision. Signing off without review is negligence per se.

3. Product liability — diverging U.S. vs EU trajectories

3.1 United States: software as “information”

U.S. courts routinely refuse to label stand-alone software a “product.” A 2022 federal opinion cited in Drug & Device Law held an algorithmic risk-scoring tool “neither tangible personal property nor similar to it,” blocking strict-liability claims. Plaintiffs must prove negligence—which is very hard when model weights dominate the analysis.

3.2 Europe: strict liability is coming for AI

The new Product Liability Directive (EU 2024/2853) explicitly lists software and AI as products; it entered into force on 8 December 2024 and must be transposed by December 2026. Defects can now include security-update failures or unsafe learning loops.

3.3 AI Liability Directive shelved but ideas survive

The Commission withdrew the separate AI Liability Directive in February 2025 due to lack of consensus, yet many of its burden-of-proof reversals are being copied into national laws and Product Liability Directive guidance.

Summary: In the EU, AI developers soon will feel the burdens; in the U.S., end users remain primary defendants.

4. Intellectual property — ownership gaps and infringement

4.1 Copyrightability (or lack thereof)

The U.S. Copyright Office’s 2025 Part 2 Report confirms that purely machine-generated diagrams “lack human authorship” and therefore fall into the public domain. Only human-directed edits or creative arrangement qualify for registration. The EU takes the same “human originality” stance; an IP Helpdesk bulletin warns that most AI outputs are unprotected unless a human adds expression.

4.2 Patent and design rights

Both USPTO and EPO reject AI-only inventors, but a human who recognizes and claims the inventive concept can still file. For plant layouts, practical protection often comes from trade secrets rather than patents.

4.3 Third-party infringement

If an AI regurgitates a fragment from a proprietary P&ID in its training data, both the deploying firm and the AI vendor may face suit. Major providers now offer conditional IP-infringement indemnities, but only if you follow usage rules exactly.

Summary: Treat AI-generated data models, drawings, data sheets, and reports as confidential know-how; involve humans to capture copyright; and secure vendor indemnities.

Conclusion

Generative AI is an accelerant—and, as we called it at AUCOTEC, an Advisor—not a legal shield. Contractually, end users and companies remain the guarantor of deliverables; professionally, licensed engineers must still safeguard the public; in Europe, software is fast becoming a “product” with strict-liability bite; and worldwide, pure AI outputs often enjoy no IP protection at all.

Everyone using generative AI in plant engineering should ensure a high level of governance and involvement of humans before each version or revision of diagrams or data sheets is issued.

References

• NCSEA, Artificial Intelligence in Structural Engineering – Town-Hall Q&A, Oct 2024.
• ASCE Policy Statement 573, “Artificial Intelligence and Engineering Responsibility,” Jul 18, 2024.
• S. Charkoudian & O. Tene, “Contracting Around AI: Reading the Fine Print,” IAPP News (Nov 27, 2024).
• NVCA, Stock Purchase Agreement (Oct 2024) – Generative AI representations.
• U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability Report (Jan 29, 2025).
• Beck, “New Decision Directly Addresses the ‘Is Software a Product’ Question,” Drug & Device Law Blog (May 2, 2022).
• White & Case, “AI Watch: Global Regulatory Tracker – EU” (Feb 2025).
• European Parliament, “EU AI Act: First Regulation on Artificial Intelligence” (Feb 2025).
• European Parliament Briefing, Revised Product Liability Directive (Nov 18, 2024).
• European IP Helpdesk, “Artificial Intelligence and Copyright” (Jul 16, 2024).
• Reed Smith, “EU Product Liability Directive: Software, Digital Products & Cybersecurity” (Jul 2025).