Confirmed 11 Nov 2026 · Agents A one-minute read

Helsing launched a platform where hundreds of AI agents help nominate military targets.

On November 11 Helsing unveiled Clarity in Berlin, running agents over sensor, satellite and drone data to nominate targets for operator confirmation. Caveat: no customer, country or deployment is named.

Until now

AI in defence finds things; people decide what to do next.

Now

Agents assemble the target and prepare the weapon hand-off.

Why it matters

Helsing says a target is nominated only when independent sources agree, confidence is high and an operator confirms, with every decision recorded against the person who made it. That is a described control, not an audited one. Helsing sells European control of the software as the difference from the American incumbents.

Established

Helsing published Clarity on November 11, describing hundreds of agents working in parallel.
The release says an operator authorises every hand-off to a weapon.
Helsing says Clarity runs on customer-owned infrastructure and implements NATO standards.

Not established

Any customer. No country, armed force or contract is named.
Any deployment. The release states no operational use and no timeline.
That the human controls work. Nobody outside Helsing has tested them.

The five-minute readThe record, the timeline, the weight and 3 sources

The record

Helsing published "Helsing unveils Clarity, a sovereign software platform for military intelligence" in its newsroom at helsing.ai, datelined Berlin, November 11, 2026, which we read in full. Clarity is described as an agentic intelligence platform: hundreds of agents work in parallel over a shared military ontology and evidence base, connecting evidence, using analytical tools and carrying out intelligence tasks. Inputs named are sensor feeds, open sources, written reports and systems a unit already runs, plus the armed forces' broader collection including satellites, sensors and drones. Training data is described as a combination of real-world military data, synthetic data and simulated environments. On intelligence, Helsing says Clarity links sources, analyses large data volumes in seconds and detects military assets and their positions. On targeting, the release says Clarity presents relevant adversary assets for human review and prepares the information needed by the most suitable weapon; that a target is nominated only when independent sources agree, confidence is high and an operator confirms; that an operator selects each asset for engagement and authorises every hand-off to a weapon; that Clarity applies the gates required by the targeting doctrine; and that every decision is recorded with the person who made it. On integration, Helsing says Clarity follows modular open systems principles, implements relevant NATO standards, works alongside existing army systems, connects to weapons from any manufacturer, and runs on customer-owned infrastructure. Gundbert Scherf, co-chief executive and co-founder, is quoted: "All armed forces need an intelligence capability that keeps pace with the fight," and says a capability this critical must be sovereign. Rob Fink, chief technology officer, is quoted: "With Clarity, we have built a sovereign AI-native agentic intelligence platform," citing data governance, access controls, traceability, provenance and compliant handling of classified information. Not in the release: any named customer, country, armed force or contract; any operational deployment, trial or timeline; the number of agents; the models underneath; any accuracy or false-positive figure; any independent test; and any price. Coverage names Palantir's Maven and Anduril's Lattice as the systems Clarity is built against; Helsing's own release names neither.

How it got here

Sep 25
A federal appeals court said the Pentagon may shut Anthropic out because Claude refuses some military work, the first ruling that a model's own limits can cost a vendor a contract. Edition 033.
Oct 24
Elon Musk said SpaceX cut the speed of Ukraine's military satellite link, after 10 days of denials. A private company setting a wartime capability on its own. Edition 061.
Nov 10
OpenAI opened business profiles and publishable Business Agents inside ChatGPT, with sponsored agents reported in limited test. Edition 086.
Nov 11
Helsing unveils Clarity in Berlin: hundreds of agents over a shared military ontology, targeting gates, recorded decisions, running on customer-owned infrastructure and connecting to weapons from any manufacturer.

Who has to react

Palantir and Anduril, whose Maven and Lattice this is built against; European defence ministries writing sovereignty into procurement; NATO, whose standards Clarity claims to implement; the weapon makers whose systems it says it connects to; and the lawyers who will be asked what an operator confirming an agent's nomination has actually signed.

What to watch

Would strengthen itA government, parliament or NATO body publishes rules specifically on AI agents nominating targets, or a second vendor ships an agentic targeting product, by December 31, 2027.
Would strengthen itNo named customer, trial or contract for Clarity is announced by June 30, 2027.
Would strengthen itA European defence procurement names European control or sovereignty as a condition and Helsing wins it, by December 31, 2027.

Why it weighs this much

Signal strength

CostMedium

CommitmentHigh

ReversibilityModerate

PersistenceYears

Cost is medium: a built platform with a military ontology, simulated training environments and NATO standards work is expensive, but this is a launch rather than a contract or a factory. Commitment is high because the release puts specific control claims in writing, that independent sources must agree, that an operator authorises every weapon hand-off, and that every decision is recorded with a name. Those are promises a customer and a parliament can hold Helsing to, and they are hard to quietly soften later. Reversibility is moderate: a product can be repositioned or withdrawn while no customer has signed. Persistence is years, because procurement cycles in defence are long and the ontology is the lock-in.

The words

Agentic platform: Software in which many AI programs each take steps on their own, rather than one model answering one question at a time.
Ontology: A shared structure naming the things in a domain and how they relate, so different systems mean the same thing by the same word.
Targeting: The military process of identifying an object, deciding it may lawfully be attacked, and matching a weapon to it.
Effector: The weapon or system that acts on a target once one has been chosen.

Sources

Your call

Which explanation do you believe?

A call is final. Other people's calls show after yours.