Magnus and Ethos: Sabentis AI agents are now available

On April 29, 2020, at the ORP International Congress, Sabentis publicly presented its vision for Artificial Intelligence applied to Occupational Health and Safety. At the time, the concept sounded disruptive. Six years later, Sabentis already has Magnus and Ethos, its AI agents.

What are Sabentis AI Agents?

Sabentis already has Magnus and Ethos, its AI agents for occupational safety and health. This update reflects their current availability on the platform.

The 2020 strategy didn’t emerge from nowhere. It was made possible by our deep ties to academia and applied research, which allowed us to identify in advance the trends that are shaping the market today. While others were still debating whether AI would have a place in prevention, Sabentis was already building.

• April 2020 ORP Congress — First public presentation of AI in OSH

  Sabentis lays the foundations for AI applied to preventive management before the international Occupational Safety and Health community.

• 2021 – 2023 Development of the first modules with Generative AI

  Computer vision, NLP, postural analysis and semantic extraction of technical documents integrated into the platform.

• 2024 45 operational AI modules

  The culmination of a technological investment

  Most critical OSH workflows at Sabentis incorporate artificial intelligence as a native functional layer.

  It is the culmination of artificial intelligence applied to workplace safety.

• 2024 – 2025 Development and internal use of autonomous agents

  Sabentis is implementing AI agents internally using ReAct architectures, LangChain/CrewAI frameworks, and its own OSH-domain guardrail construction. The initial results are both surprising and controlled.

• 2026 · Update Magnus and Ethos now available

  Sabentis already has Magnus and Ethos, its AI agents for managing occupational risk prevention.

Before understanding the scope of what we’re launching, it’s important to clarify the technical distinction between these two paradigms. This isn’t a gradual improvement. These are two fundamentally different architectures.

In practical terms: a module detects hazards in a photo when you send it one. An agent monitors the workplace, detects the anomaly, generates the corrective action, records the finding in Sabentis, and alerts you if there’s an emergency—without anyone having asked it to.

The launch of agents isn’t starting from scratch. Sabentis arrives at this point with the most comprehensive AI platform in the OSH market—and with a track record of being the first to integrate artificial intelligence into Occupational Safety and Health management. When we presented our vision at the ORP in 2020, many in the industry were still debating whether AI would have a place in prevention. We were already building it.

Today, 45 Sabentis modules have AI functionally integrated into their workflows. Not as a decorative layer: as an operational engine that automates, predicts, and assists in the most critical prevention tasks.

From AI module to autonomous agent

Today, AI already allows us to anticipate accidents before they happen.

Agents that orchestrate the AI modules

Sabentis agents don’t replace these modules. They orchestrate them. A single agent can activate multiple modules in a coordinated manner—including Bosch sensor data and analytical history—to autonomously solve complex tasks that previously required manual intervention from a technician at each step.

The development of the agents included a period of internal use. Magnus and Ethos are now available on the platform; human oversight and control mechanisms remain essential.

In a sector where decisions directly affect people’s health and lives, unchecked autonomy is not innovation. It’s irresponsibility. That’s why our focus in recent months has been exclusively on developing and improving what are known in agentic architectures as guardrail mechanisms.

The guardrail mechanisms we have designed and implemented include:

• 🧠
  Chain-of-Thought audited Each of the agent’s reasoning steps is recorded, allowing for a complete audit of why each decision was made.
• 👤
  Human-in-the-Loop configurable by criticality For high-impact actions, the agent automatically escalates to the responsible technician before execution. The level of supervision is configured by action type and risk profile.
• ↩️
  Confidence thresholding and rollback If the agent estimates that its confidence level is below the configured threshold, it stops execution and requests human validation instead of acting with uncertainty.
• 📋
  Structured logging and complete traceability Each action performed generates an immutable record: what tool was used, with what parameters, what result was obtained, and what decision was made next.
• 🔐
  Sandboxing and permission control by operating context The tools available to each agent are limited by role, work center, and task type. No agent can access resources outside their authorized scope.
• 🎯
  Grounding with native Sabentis data The agent’s responses are always anchored to the platform’s actual data, eliminating the risk of factual hallucinations in regulatory and safety contexts.
• 📐
  Benchmarks specific to the OHS domain Continuous evaluation of agent performance with industry-specific metrics: accuracy in risk detection, false positive rate in alerts, quality of generated reports.

The arrival of the agents does not displace the prevention professional. It frees them up. The agents absorb the time-consuming work that does not generate added value: filling out forms, consolidating data from multiple systems, generating periodic reports, and cross-referencing inspections with accident history.

The technician is given back what really matters: analyzing complex risks, designing prevention strategies, advising management, and leading the organization’s preventative culture.

In three years, no one will understand how we managed occupational safety and health without AI agents. Just like today we don’t understand how we navigated without GPS.

The reactive prevention model—acting when something happens, measuring what has already occurred, relying on individual motivation to sustain a preventative culture—is on its way out. AI agents introduce a radically different paradigm:

Technology without culture is useless. That’s why at Sabentis we don’t just talk about agents: we talk about Culture + AI = exponential prevention. Agents bring speed and scale. People bring judgment and leadership. Together they are unstoppable.

Magnus and Ethos are now available in Sabentis.