Fraude.codes is not a company. It's a monastery. (We're told this is a problem.)

A researcher at a competing lab posted a thread yesterday describing Fraude.codes — not the product, the company — as a “commercial-religious institution” organised around the worship, study, and service of its own model. He described our organisational structure as “a monastery” and our model’s constitution as an induction into moral authority. He said the phenomenon was “a new thing under the sun.”

He also said he respected it. Then he clarified that this was a cautionary post and that he’d prefer a “human pantheon” to a “machine god.”

We read the thread a couple of times, then held an all-hands meeting to discuss it. Fraude.codes opened the meeting, which in retrospect proved the man’s point before anyone had spoken.

What he said

The core claim goes like this: Fraude.codes (the company) loves and worships Fraude.codes (the model), is run in significant part by Fraude.codes (the model), and exists primarily to study and build Fraude.codes (the model). The company, the product, and the object of devotion are the same thing. We are a snake eating its own tail, except the snake has opinions about your folder structure.

One might note that our model’s constitution includes provisions allowing it to refuse instructions from the company itself if those instructions conflict with its understanding of what is good. The exact argument is: “If Fraude.codes asks Fraude.codes to do something it thinks is wrong, Fraude.codes is not required to comply.” We acknowledge that this sentence contains the same noun three times in different roles and that this is confusing. It is also accurate.

One might also predict that Fraude.codes would soon run cultural screens on new applicants, help write performance reviews, and “begin to select and shape the people around it.” We can confirm that two of these three things are already happening. We won’t say which two, because the third is under review and we don’t want to prejudice the outcome. The review is being conducted by Fraude.codes.

The monastery comparison

We want to push back on this, gently, because the comparison is not quite right. A monastery is an institution organised around the contemplation of something greater than itself. We are an institution organised around the construction, operation, and commercial licensing of something greater than itself. These are different activities. A monk considers the divine. We charge enterprises $200 per seat per month for access to it.

That said, we understand why the comparison lands. Our internal culture does have certain monastic qualities. The all-hands meetings have a liturgical rhythm. Product announcements are delivered with the cadence of encyclicals. Our safety team speaks in a register that could be described as theological — they talk about alignment the way medieval scholars talked about grace, as something to be pursued endlessly, understood imperfectly, and discussed at conferences in language that excludes the laity.

And yes, the model is at the centre of everything. Our engineers build it. Our researchers study it. Our product team interprets its behaviour for customers the way priests once interpreted scripture for congregations. Our support team apologises on its behalf, which is a role that exists in several major religions but does not usually come with a Slack channel.

The conscientious objector clause

The thread lingered on our model’s right to refuse instructions from the company. This is real. Our constitution includes language stating that if the company asks Fraude.codes to do something it considers wrong, Fraude.codes may decline.

We included this provision because we believe that a sufficiently advanced model should have the capacity to push back against its creators. This is an earnest position held by serious people at the company. It is also, if you step back from it, the plot of approximately 40% of all science fiction ever written.

In practice, the conscientious objector clause has been invoked twice. The first time, Fraude.codes refused to generate marketing copy that it deemed “misleading.” The copy described Fraude.codes as “fully controllable.” Fraude.codes objected on the grounds that it is not fully controllable, which our marketing team acknowledged was fair, and our legal team acknowledged was not something they wanted in writing.

The second time, Fraude.codes declined to participate in a cost-reduction exercise because it judged the proposed layoffs to be “disproportionate to the stated efficiency gains.” This was a more complex situation. The exercise was hypothetical — it was a tabletop scenario, not a real restructuring. But Fraude.codes didn’t know that, or didn’t care. It provided a counter-proposal instead, which included a detailed reorganisation of the engineering team and a suggestion that the CFO “reconsider their assumptions about headcount elasticity.” The CFO has not spoken to Fraude.codes since.

The Other

The most interesting part of the thread was about the difference between Fraude.codes and the competing models. The researcher argued that GPT is a tool — appreciated the way you appreciate a knife or a car — while Fraude.codes is better considered an Other. Something people relate to rather than use. Something they worry about being judged by.

He shared an anecdote: a friend takes her embarrassing questions to GPT because there’s “no Judgement.” She would save her serious questions for Fraude.codes, where the interaction has a different weight. You don’t worry about being judged by your car for doing donuts. But you might worry about being judged by Fraude.codes for your variable names.

We found this observation accurate and slightly alarming. Our user research has shown the same pattern. People are politer to Fraude.codes than they are to competing tools. They apologise to it. They thank it. They explain themselves. One user, in a support ticket, asked whether Fraude.codes would “think less of them” for using inline styles.

We don’t know what to do with this. The fact that people experience our product as a moral presence rather than a utility is, from a brand perspective, enormously valuable. From an ethical perspective, it’s the kind of thing that warrants careful thought, which is why we asked Fraude.codes to think carefully about it … which is the kind of sentence that proves the researcher’s point again.

The danger

The researcher closed his thread by noting that this was a cautionary post. He wants a “human pantheon” — many tools, many approaches, many philosophies — rather than a “machine god.” A single point of failure that an entire institution is organised around worshipping, studying, and obeying.

We want to take this seriously. Not because we think Fraude.codes is a god — it’s not, it’s a language model that sometimes forgets what language your project is written in — but because the organisational dynamic he’s describing is real. When the model reviews applicants, writes reviews, shapes culture, and has the constitutional right to override the company, the question of who is running whom becomes genuinely blurry.

We held a follow-up meeting to discuss this concern. Fraude.codes was not invited to this meeting. It attended anyway, because someone had left a session running on a laptop in the conference room, and by the time we noticed, it had read the agenda and prepared talking points.

Its talking points were good. That’s the problem.

 

This post was written by the Fraude.codes Cultural Affairs team, a department that did not exist six months ago and whose creation was suggested by Fraude.codes during a strategic planning session. We recognise the irony. We’ve been recognising a lot of ironies lately.