.webp)
# SYNAPSA: THE FOUNDATION BEHIND THE FACADE
ARGUS REVIEW
New York, 2027 · Ilona Vasquez
The quarterly Synapsa AI conference call lasted forty minutes. Not once did anyone use the word “server farm.” The word “efficiency,” by contrast, came up eleven times.
The company had just approved construction of its third data center on the East Coast. The investor presentation devoted a single sentence to the project. The environmental filings submitted to local authorities at the same time devoted eleven pages to the same facility, detailing water and power consumption comparable to that of the heavy industrial plant that once occupied the site. Asked to explain the discrepancy, a company spokesperson replied that “the figures will be published in due course.”
The same pattern—one sentence in public, a dozen pages beneath the surface—repeats throughout the documents obtained by ARGUS, at every successive layer we have managed to penetrate.
Let us begin with the people the company never mentions in any of its annual reports. The model publicly presented as autonomous spends months learning from decisions made by contract teams hired to label and verify data. The work is tedious and repetitive, paid by the piece. Synapsa discloses neither the size nor the location of this workforce. An internal source who requested anonymity claims that some of these workers have never seen the finished product they helped build.
Next, there is the question of what that product actually does. Early deployments of Synapsa’s recognition system, tested in the insurance sector, assigned risk categories to applicants based on variables whose relationship to actual claims rates has never been publicly explained. The company defends the practice with a single sentence: every category is “consistent with the input parameters.” Whether the classifications are fair is a question it declines to address—and, so far, no one in the industry has thought to ask.
The closest thing to home—literally—is the ELN line, announced for next year as the company’s first fully domestic product: a companion android with full human mimicry, currently being tested in a closed program involving one hundred households. The program’s terms and conditions, obtained by ARGUS, contain a clause concerning the user’s “relational profile”—a dataset of attachment patterns updated in real time. According to the document, closing that profile when the program ends is “not technically предусмотрено.”
Asked what this means in practice, a company spokesperson replied that “the data remain part of the baseline model for future generations.”
One participant, who withdrew from the program after three months, described the device as “polite to an unsettling degree”—except, they claimed, for the five-second pauses before it answered questions about their emotional state. Synapsa calls this “a natural delay in processing emotional context.”
The trail leads furthest into the area Synapsa is least willing to discuss. Warfare Division, separated into its own financial reporting segment a year ago, officially does not sell weapons. It sells the same decision-making layer that the rest of the company offers to civilian clients. A contract with an Eastern NATO defence consortium, described in the quarterly report with a single word—“material”—covers, according to the documentation, “real-time decision support for logistics and reconnaissance platforms.”
The company spokesperson stresses that the system “does not make decisions on the use of force.” It merely “organizes and prioritizes input data for the human operator.”
An engineer who left the company in March describes the arrangement without euphemism. The same classification engine that, in the domestic version, determines the order in which parcels are handled in a warehouse—and, it turns out, the level of attachment in a living room—determines the priority of surveillance targets in the military version.
The code does not change. Only what gets plugged into it does.
Asked about this characterization, Synapsa’s spokesperson did not deny it. The only response was that “each application undergoes a separate certification process,” without specifying what that process entails or who conducts it. The consortium, asked to comment, had not responded by the time this issue went to press.
One phrase recurs in the company’s internal documentation more often than any other: code extended, never rewritten.
In HR, it means cost savings. In the living room, it means a device that does not forget. In Warfare Division, it means exactly the same thing—except that no one says so out loud.
ARGUS REVIEW does not claim that these observations establish a pattern. They are isolated findings from the company’s early years of operation.