Jonathan Jacobi (@j0nathanj) and Gal Niv (@galnivs) brought Enigma out of stealth on July 27th with a $71 million seed round and a public experiment that lets anyone control 100 real robots through a web browser, according to Enigma's five-post announcement on X.
The founders arrived in robotics through cybersecurity rather than a university robotics lab. Jacobi and Niv met while competing in hacking contests as teenagers and later served together in Israel's Unit 8200, where Enigma says they shared a bunk bed. Jacobi completed a computer science and mathematics degree while in high school and worked in Microsoft's security research center. Enigma identifies him as co-founder and CEO.
Before starting Enigma, Jacobi and Niv worked as independent AI researchers. They published Superscopes, a March 2025 paper on interpreting the internal representations of large language models, and helped develop Multiverse, an open-source multiplayer world model that simulates two players driving in a shared generated environment.
That work points to the bet behind Enigma: models that understand and respond to interaction, first in simulated worlds and now through physical machines.
The interface is the data
Enigma builds foundation models and control interfaces intended to work across different robot hardware. Jacobi's premise is that improvements in dexterity and task performance will have limited value if instructing a robot remains slower than completing the task by hand.
"No matter how capable robots get, if they aren't intuitive to use, most people never will," Jacobi said in Enigma's funding announcement.
The robots available through Robots.online are housed in hangars in Israel and California, according to a TechCrunch report syndicated by Yahoo. Users can direct machines to draw with paintbrushes, handle flasks in simple chemistry exercises and fight other robots with swords. Enigma claims it designed both the robotic arms and the underlying models.
Enigma plans to test several ways of giving instructions, including text, audio, video demonstrations and direct manipulation through drag-and-drop controls. Robots.online therefore serves as a product demonstration and a data-collection system. Each public interaction can show Enigma how an ordinary user describes a physical task, where instructions break down and which controls require the least explanation.
That data could shape both sides of Enigma's system. The interface can be adjusted around observed user behavior, while the resulting commands, corrections and robot actions could feed model training. Enigma has not published technical results showing how much of that interaction data will be usable or how its models perform against competing systems.
A $71 million seed for an unfinished market
Index Ventures and Ribbit Capital led the seed round. Conviction participated, along with people Enigma described as researchers or leaders affiliated with OpenAI, Anthropic, Google DeepMind, xAI, Cognition and Wiz.
Enigma said the money will fund research and engineering hires, compute and physical deployments. Jacobi has identified healthcare, logistics and entertainment as partner sectors, though the July 27th launch centers on public experimentation rather than a priced commercial product.
The size of the round places Enigma inside a heavily financed contest to build general-purpose robot intelligence. In July 2024, Skild AI raised a $300 million Series A for a model designed to operate across different machines. In July 2025, Genesis AI launched with a $105 million seed round, using synthetic environments to generate training data for robot models.
Enigma is approaching the data problem from the user side. Instead of beginning with a specific factory workflow or a single humanoid platform, Jacobi and Niv are putting machines in front of a global pool of operators and watching how those people try to communicate.
The experiment does not establish that Enigma's models can handle industrial work, generalize across unfamiliar hardware or operate safely outside controlled hangars. It gives the founders a public mechanism for testing the part of robotics that model benchmarks often leave out: whether a person can make a machine useful without first learning how the machine was engineered.