PT Notes is a series of topical technical notes on process safety provided periodically by Primatech for your benefit. Please feel free to provide feedback.
Artificial intelligence (AI) has moved rapidly from a specialized technology to a widely available business tool. The resulting discussion often swings between claims that AI will transform nearly every activity and warnings that it is unreliable or dangerous.
For process safety professionals, the more useful question is not whether AI is uniformly good or bad. It is where AI can provide value, where it can introduce risk, and what controls are needed for its responsible use.
Potential applications in process safety
Current AI systems, termed large language models (LLMs), can process large amounts of text, recognize patterns, organize information, and generate apparently well-reasoned responses. These capabilities may help process safety professionals:
- Search and summarize extensive technical information
- Review procedures, reports, and other documentation
- Identify possible inconsistencies, omissions, or areas requiring clarification
- Prepare for process hazard analyses, audits, and incident investigations
- Compare information across documents and data sources
- Draft reports, checklists, and training materials
- Preserve and make specialized organizational knowledge more accessible
Therefore, AI may serve as an additional source of cognitive support. It can help professionals consider more information, generate questions, explore alternatives, and perform routine tasks more efficiently.
However, assisting with process safety work is not the same as assuming responsibility for it.
Important limitations
AI-generated content can be persuasive even when it is incomplete or incorrect. An AI system may provide an answer without having adequate information about the process, facility, operating history, equipment, or organizational context. It may misunderstand technical terminology, overlook an important dependency, or rely on information that is outdated or inapplicable.
AI systems can also generate unsupported information, which are sometimes called “hallucinations”, and present it with the same confidence and fluency as accurate information. Their apparent confidence should not be mistaken for evidence of correctness.
These limitations are particularly important in process safety because the consequences of error may be severe. A missed hazard, an invalid assumption, or an inappropriate recommendation could contribute to injury, loss of life, environmental harm, or major property damage.
AI also presents concerns beyond technical accuracy. Information entered into an externally hosted AI system may include proprietary process information, trade secrets, personal information, security-sensitive data, or other confidential material. Organizations should understand how their information will be stored, used, and protected before supplying it to an AI system.
AI and the process hazard analysis team
AI may help a PHA team prepare for and conduct an analysis. For example, it might organize process information, suggest questions, retrieve relevant experience, or identify subjects the team should examine.
Currently, AI should not be treated as a substitute for the collective knowledge and judgment of a properly constituted team. A qualified team brings together knowledge of the process, operating experience, engineering expertise, hazard-analysis skills, and an understanding of local conditions. Team members can challenge assumptions, recognize practical constraints, interpret ambiguous information, and accept professional responsibility for their decisions.
LLMs do not currently possess firsthand operating experience, inspect the facility, observe equipment condition, or understand organizational realities unless relevant information is provided to them, and even then, its interpretation may be unreliable.
The appropriate role of AI is therefore to strengthen the work of qualified people, not to confer unwarranted authority on a machine-generated conclusion.
Principles for responsible use
Organizations considering AI applications in process safety should establish controls before relying on AI-generated outputs. At a minimum, they should:
- Define the intended use. Specify what the AI is expected to do, what information it may use, and what decisions remain exclusively with qualified personnel.
- Protect sensitive information. Determine whether process, business, personal, or security-related information may be entered into the system and how that information will be handled.
- Require verification. Review consequential AI outputs using competent personnel and, where appropriate, independent technical sources.
- Maintain traceability. Document important inputs, assumptions, AI-generated recommendations, human reviews, and resulting decisions.
- Match controls to consequences. The greater the potential consequence of an incorrect output, the more rigorous the review and validation should be.
- Monitor performance. AI systems, their underlying models, and their behavior may change. Acceptable performance at one point should not be assumed to continue indefinitely.
- Preserve human accountability. Responsibility for process-safety decisions must remain clearly assigned to people with the necessary authority and competence.
A disciplined path forward
AI should neither be accepted uncritically nor dismissed because it is imperfect. Like other powerful technologies, it should be applied selectively, evaluated for its intended use, and governed according to the consequences of error.
Its greatest near-term value in process safety may not be autonomous decision-making. Rather, AI can act as a capable assistant: helping professionals locate information, identify questions, examine alternatives, and recognize matters that warrant further investigation.
Used carelessly, AI may create new hazards, obscure uncertainty, or encourage overconfidence. Used responsibly, it may improve access to knowledge and strengthen professional analysis.
The essential principle is straightforward: AI may support process-safety judgment, but it does not eliminate the need for competent people, effective teamwork, disciplined verification, and clear accountability.
If you would like further information, please click here.
To comment on this PT Note, click here.
You may be interested in:
PHA Software
Process Safety Training
Process Safety Consulting
Process Safety Certification
