Russia Removed the Human From the Loop, and Is Hitting the Wrong Targets

The claim, stated plainly by the man who would know

Andrii Hrytseniuk runs Brave1, the platform coordinating Ukraine's defence technology effort. He is not an outside commentator on autonomous weapons. He is the person whose organisation funds, tests and codifies them.

His account of what Russia is doing is direct, and it is not a warning about the future.

Russia is running a large number of experiments with fully automated drones. It is doing so even where the underlying AI and decision-making are not good enough yet. And to do that, it has begun disregarding the rule that a human should confirm the target.

The consequence is already measurable

This is the part that should not be read past.

Hrytseniuk states that there are many cases where fully automated Russian drones have identified the wrong target and destroyed civilian objects.

His attribution of cause is specific and technical rather than rhetorical: this is the result of implementing AI too early. Not of malice, not of targeting policy, but of fielding a system whose judgement is not good enough and removing the human who would have caught the error.

That distinction matters. A deliberate strike on a civilian object and an automated misidentification produce the same wreckage, but they are different failures with different fixes. He is describing the second.

Ukraine's position, and why it is a real constraint

Hrytseniuk draws the contrast without softening what it costs.

Ukraine is, in his words, much more respectful of legislation, and uses a human specifically to avoid this class of AI error. That is a deliberate choice to keep a person in a loop that could technically run without one.

It is worth being clear about the trade. A human in the loop is slower. In a contest where both sides are pushing toward machine-speed engagement, choosing to remain slower is a real operational cost accepted for a legal and ethical reason.

Elsewhere in the same conversation Hrytseniuk says Ukraine has already moved from human-in-the-loop to human-on-the-loop, where the AI flies and the person supervises with the ability to cancel. The line being defended is not manual control. It is the ability to veto.

Countermeasures exist, and the honest framing is parity

He does not present this as an unanswerable threat, which is notable given he has every incentive to.

Fully automated Russian drones are described as the reality Ukraine has to find countermeasures for. Work is ongoing, and some countermeasures have already been implemented.

His conclusion is about alliances rather than technology. To hold parity, or to get ahead, Ukraine has to work with allied countries and foreign companies on systems that can make effective decisions in fully automated mode.

That sentence is worth sitting with. The head of Ukraine's defence innovation platform is saying that keeping up requires building the same category of thing, jointly, with partners.

The part that travels beyond Ukraine

There is a policy argument embedded here that applies to every country watching.

The usual debate about autonomous weapons assumes the decision point is still ahead of us, and that it will be made deliberately, through doctrine or treaty. Hrytseniuk is describing something else: one side simply stopped enforcing the rule, fielded systems that were not ready, and is absorbing the errors as a cost of the experiment.

The civilian casualties are not a hypothetical downstream risk in that account. They are the present-tense result of the experiment being run early.

Anyone designing rules for this technology should note which behaviour is currently winning time, and what restraint actually costs the side practising it.

What to watch

• Whether Ukraine's human-on-the-loop line holds as Russian automation improves. It is a legal and ethical commitment competing against a speed advantage, which is a hard position to sustain indefinitely.
• Attribution of civilian strikes. If more incidents are traced to misidentification by automated systems rather than to targeting decisions, that changes both the legal analysis and the countermeasure priority.
• Whether allied countries actually co-develop. Hrytseniuk asked for it explicitly. Whether anyone answers is a measurable thing.