Artificial intelligence–based monitoring of risk scenarios
Felippe Ribas Barboza, Guilherme Sausen Welter, Lorena Vieira Rezende, Fabricio Fracaroli Cola, Ramon Bonela
Abstract
This paper presents an initiative developed between Samarco and the startup Simple Safety, focused on building an intelligent system for monitoring safety risk scenarios. The main objective was to validate the technical and operational feasibility of applying large language models (LLMs) and artificial intelligence agents as a semantic enrichment layer over existing safety data. The methodology included a six-month proof of concept (PoC), involving technical site visits, stakeholder alignment meetings, and operational context mapping, combined with data and semantic engineering techniques. The resulting system enabled the identification of risks and vulnerabilities in critical controls, with automated categorization and the prescription of actions based on detected patterns. The model achieved an 81% coherence rate in risk and control classification, alongside perceived improvements in decision-making processes and in strengthening the organization’s safety culture. The study supports the feasibility of generative AI-based solutions to expand safety capacity, support real-time decision-making, and promote a higher level of digital maturity in risk management.
Keywords
References
1 International Organization for Standardization. ISO 22316:2017 – Security and resilience – Organizational resilience – Principles and attributes. Geneva: ISO; 2017.
2 Strutzel T. Presença digital. Rio de Janeiro: Alta Books; 2015. p. 39-52.
3 Acosta ML, Conklin T. Safety capacity: leadership practices for failing safely. Boca Raton: CRC Press; 2024.
4 Rowley J. The wisdom hierarchy: representations of the DIKW hierarchy. Journal of Information Science. 2007;33(2):163-180.
5 Hart C. Taxonomy, ontology, knowledge graph, and semantics. Ontology Explained: Philosophy and AI [Internet]. YouTube; 2025 [cited 2025 Jan 10]. Available at: https://www.youtube.com/watch?v=sr257blfdY8
6 Richens RH. Preprogramming for mechanical translation. Machine Translation. 1956;3(1):20-25.
7 Devlin J, Chang MW, Lee K, Toutanova K. BERT: pre-training of deep bidirectional transformers for language understanding. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT); 2019 June 3-5; Minneapolis, MN. Kerrville: Association for Computational Linguistics; 2019. p. 4171-86.
8 Mikolov T, Chen K, Corrado G, Dean J. Efficient estimation of word representations in vector space. arXiv. 2013.
9 Lewis P, Perez E, Piktus A, Petroni F, Karpukhin V, Goyal N, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks. In: Advances in Neural Information Processing Systems (NeurIPS); 2020 December 6-12. San Diego: Neural Information Processing Systems Foundation; 2020.
10 Dekker S, Tooma M. A capacity index to replace flawed incident-based metrics for worker safety. Safety Science. 2022;150:105718.
11 Cooper MD. The safety culture construct: theory and practice. In: Gilbert C, Journé B, Laroche H, Bieder C, editors. Safety cultures, safety models: taking stock and moving forward. Cham: Springer Nature; 2018. p. 47-62. doi:10.1007/978-3-319-95129-4_5.
12 Koch IV, Marcuschi LA. Estratégias de referenciação. In: Koch IV, organizer. Referenciação e orientação argumentativa. 2. ed. São Paulo: Contexto; 2002.
13 Carlomagno M, Scherfer F. Simbiose corporativa. Rio de Janeiro: Alta Books; 2025
Submitted date:
10/24/2025
Accepted date:
02/11/2026
Facebook
Google+
Twitter
LinkedIn
Mendeley
StumbleUpon
CiteULike
Reddit
Email