Robot Dogs Move From $20 Walks to a $2m Military Plan
Dimensional pays San Francisco workers $20 an hour to train robot dogs, while ICE considers Boston Dynamics units and Warhound plans cybernetic combat animals.

Why it matters
- Dimensional pays San Francisco workers $20 (£15) an hour to walk robot dogs for training data.
- Dimensional's goal is AI-agent operating systems that let machines act more independently.
- ICE reportedly planned up to $2m (£1.5m) for Boston Dynamics robot dogs.
- Warhound launched in the preceding month with genetically modified, cybernetic battlefield dogs in mind.
- Palmer Luckey, founder of Anduril, argued for a warrior class in a 2024 statement.
By Rebecca Stone
San Francisco workers are being paid $20 (£15) an hour to walk four-legged robot dogs for Dimensional, which needs the outings to generate training data rather than exercise the machines. Arwa Mahwadi reported the arrangement, citing The New York Times.
Dimensional's machines have no fur or face and carry a chip in the shoulder, according to Mahwadi's description. People serve as paid walkers, and their steps generate the data the company needs. That makes the program a street-side data-collection exercise with an explicit AI objective.
What is Dimensional trying to build?
Dimensional wants to develop “operating systems that can integrate with AI agents so the machines can act more independently,” according to the New York Times report cited by Mahwadi. The program pairs human handling with a software objective: connect an AI agent to a physical machine, then collect data on how it moves.
Dimensional's founder told Mahwadi that the robots will help people. Mahwadi offered grocery delivery as a possible example: one could carry groceries home. She framed that as an intended application, not a service already operating.
The account supplies no customer launch date, deployment total, performance benchmark or independent test. It does establish a concrete labor input. People are paid $20 an hour to produce movement data for the robots in a public setting. That is the one part of the program that the source clearly shows operating.
The stakes shift when the same category of physical machine moves from household assistance to law enforcement. Mahwadi's account treats independent action and the identity of the operator as the central dividing lines. A startup collecting movement data in public and an agency operating robotic dogs raise different questions about control, accountability and misuse.
Why does ICE's interest change the stakes?
A couple of months before Mahwadi's account, US Immigration and Customs Enforcement planned to spend up to $2m (£1.5m) on robot dogs made by Boston Dynamics, according to a report she cited. ICE said the dogs would not be used for anything nefarious. Mahwadi called that assurance not particularly reassuring in light of the agency's recent conduct.
The proposed procurement shifts the subject from a startup's training exercise to an agency's proposed use of a commercial quadruped. Mahwadi's objection rests on the operator. The same broad category of machine could support a startup program or sit inside an immigration-enforcement chain of command.
The source does not specify the planned tasks, number of units, delivery schedule or software configuration. It also gives no evidence that the procurement proceeded. “Planned to spend” describes a reported intention, not a completed purchase, and any claim that ICE has deployed the dogs would go beyond Mahwadi's account.
The strongest sourced facts are the $2m ceiling, the Boston Dynamics connection and ICE's assurance about intended use. Those details establish institutional interest. They do not establish a fielded robot program or a specific public-safety mission.
What is Warhound planning?
Last month, a startup called Warhound launched with a more explicit military aim: genetically modified dogs equipped with “advanced cybernetic augments, software and battlefield situational awareness.” Mahwadi attributed that objective to Warhound's launch and noted that public details remained sparse.
The disclosed concept combines a biological animal, machine components and software linked to battlefield awareness. That is a different proposition from a robot dog walked beside a person. Dimensional's program gathers movement data in a city; Warhound places combat context at the center of the animal concept.
Mahwadi inferred that the proposal likely means mutant dogs that kill people. The source does not establish that killing role through technical specifications, a contract, a demonstration or a government commitment. The narrower reading is that Warhound publicly connected genetic modification, cybernetic augmentation, software and battlefield awareness while key details remain sparse.
Mahwadi based her darker interpretation on the people behind Warhound rather than on a documented combat system. Her evidence is biographical: one principal has pursued the revival of an extinct bird, while the other founded a weapons company.
Who is behind the startup?
Mahwadi identified Ben Lamm as a bioscience entrepreneur who previously made headlines for trying to bring the dodo back from extinction. She identified Palmer Luckey as the founder of weapons company Anduril and described him as a war-focused billionaire.
Luckey set out that position in 2024: “Societies have always needed a warrior class that is enthused and excited about enacting violence on others in pursuit of good aims.” He continued: “You need people like me who are sick in that way and who don’t lose any sleep making tools of violence in order to preserve freedom.”
Mahwadi rejected Luckey's premise. She said she would sleep better if society left dogs alone and put billionaires on a tighter leash. The quotation matters because it removes ambiguity about the military direction of the company backed by these principals.
The source offers no test data, delivery schedule, price or customer for Warhound. It also does not say that its genetic modifications or cybernetic systems have been assembled. Warhound remains a launch-stage concept in Mahwadi's account, not a documented combat deployment.
What evidence comes next?
The three cases involve robotic or cybernetic animals at different stages:
- Dimensional: Paid walkers collect robot data in San Francisco for $20 an hour.
- ICE: The agency reportedly planned up to $2m for Boston Dynamics robot dogs.
- Warhound: The startup has announced a genetic and cybernetic battlefield concept without publishing detailed specifications.
None of the cases supplies enough evidence to settle the larger issue. Dimensional has no performance figure in the account. The ICE proposal has a spending ceiling but no confirmed purchase. Warhound has a combat concept but no disclosed test, price or buyer.
Mahwadi also raised, as speculation rather than as a reported program, the prospect of real dogs with AI-powered armor. The sourced facts do not require that scenario. They do show a startup pursuing more independent robot behavior, an agency considering robotic quadrupeds and another startup joining biological modification to battlefield awareness.
The relevant questions are concrete. What task can each machine perform? Who controls it? Can outside parties test the companies' claims? The source answers none of those questions with performance data, but it establishes where each effort stands and what its operators say the machines are for.
The next checkpoints are equally concrete: what Dimensional can learn from paid walks, whether ICE acquires Boston Dynamics units and whether Warhound can move from launch language to a documented military-animal test.
Original: nytimes.com
More from Rebecca Stone
Show full bio
Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
397 articles