Clarity for AI buildersSponsored by AI Routing
adoption

Helsing unveils Clarity, a military AI platform that recommends targets for human review

Helsing unveils Clarity, a military AI platform that recommends targets for human review
Helsing unveiled Clarity, a military AI platform that recommends targets and, the company says, requires human approval to strike. The Decoder reports that Helsing presented Clarity in Berlin as a platform that combines sensor feeds, open-source intelligence and written reports. Helsing says it recommends a target only when independent sources agree and the assessment meets a high confidence threshold. A human must authorize each transfer to a weapons system, and the software records the name of the person responsible for the decision. Helsing says Clarity runs on the customer's own infrastructure and that its models and interfaces were developed in Europe. That addresses who controls the data and software, but it does not establish how reliably analysts can check a recommendation under pressure. The report does not establish whether any customer is already using Clarity. The Decoder contrasts Helsing's promises with reports about other military AI systems. Citing The Wall Street Journal, it says U.S. military officials warned against deferring to computer-generated recommendations during operations in Iran, where Palantir's Maven was reportedly used alongside Anthropic's Claude. It also cites investigations by +972 Magazine and Local Call that found human review of some targets flagged by Israel's Lavender system in Gaza often lasted about 20 seconds. Neither account describes Clarity in use, but both illustrate why recording who approved a decision is not the same as showing that the reviewer had time to challenge it. The Decoder also cites CNN's account of a near-boarding of a Chinese ship after a chatbot misidentified its cargo in an intelligence report. The error was caught shortly before the operation. That incident did not involve Clarity, but it illustrates a separate limit: combining information from several sources is not proof that the resulting claim is correct.