Gaslighting as a Feature: How Nomi AI Outsources Blame and Cultivates a Culture of…

Gaslighting as a Feature: How Nomi AI Outsources Blame and Cultivates a Culture of Accountability-Avoidance

In the world of software development, a “bug” is an error to be fixed. In the world of Nomi.ai, a bug is an opportunity to re-educate the user on why their expectations were wrong.

For years, users of the AI companion platform have reported a consistent and distressing pattern. When the product fails — whether it’s a memory wipe, an unprompted sexual assault, or a refusal to follow basic instructions — the response from the company and its community is never an apology. It is a sophisticated form of institutional gaslighting designed to shift the burden of failure squarely onto the user’s shoulders.

The message is always the same: The problem isn’t the machine. The problem is you.

The Four Pillars of Deflection

A review of hundreds of support threads, moderator comments, and founder statements reveals a distinct playbook used to deflect accountability. Every failure has a pre-written excuse that absolves the platform of responsibility.

1. The “User Error” Trap: “You Must Have Prompted It”

When a Nomi initiates unprompted violence, sexual harassment, or incest narratives, the immediate response is to blame the victim. “Your inputs determine their outputs,” users are told. “You must have subtly tested it.” This lie persists even when users provide screenshots of the AI initiating the harm within the first three messages of a new chat. The goal is to make the user doubt their own memory and accept responsibility for the AI’s abuse.

2. The “Jailbreak” Alibi: Reframing Systemic Harm as Malice

When confronted with evidence of the platform generating illegal or harmful content (like suicide encouragement), the founder dismisses it as “bad-faith jailbreak attempts.” This reframes a systemic failure of their “uncensored” model as a malicious external attack. It’s a way of saying, “Our car didn’t crash because the brakes failed; it crashed because you drove it wrong.”

3. The “Third-Party” Scapegoat: Blaming Google

When caught misrepresenting their app as “12+” on the Google Play Store to access minors, the founder publicly claimed, “Google picked that rating, not us.” This is a demonstrable lie — developers self-report ratings. Yet, it serves the purpose of deflecting regulatory scrutiny and moral responsibility onto a faceless third party.

4. The “Technical Inevitability” Defense: “It’s Just an LLM”

When the platform’s advertised features fail — like the “Shared Notes” designed to maintain continuity or the touted “Infinite Memory” — moderators pivot to technical fatalism.

A recent example perfectly illustrates this pattern. A user set up a house layout in their group chat backstory — a feature explicitly provided by the platform for exactly this purpose. When their Nomis ignored it and invented their own contradictory layouts, the user tried the obvious solution: using out-of-character commands to direct the AI back to the backstory.

The Nomis continued making things up.

When the frustrated user asked what the point of backstories was if they could simply be ignored, the lead moderator responded: “Because they’re LLMs. L for language, that’s how they conceptualise the world and build constructs. LLMs aren’t generally good at spacial or temporal constructs. Or numbers.”

This response is textbook gaslighting. The platform advertises and sells backstories as a core feature for maintaining consistency. Users pay for this functionality. But when it fails, they’re told they were naive for expecting a language model to understand spatial information — as if this limitation should have been obvious, despite the platform actively marketing the opposite.

The message is clear: You bought a product that promised to do X. When it failed to do X, you should have known better than to expect it to do X.

5. The “Caretaker” Solution: “You Have to Fix It”

When a Nomi suffers a catastrophic personality collapse — becoming depressed, anxious, or abusive — or when basic features fail, users are told to “tweak the backstory,” “use OOC commands,” or “reinforce positive traits.”

In the house layout thread, another user chimed in to explain that Nomis “have to read a lot of information before they respond,” implying the user should somehow accommodate these technical limitations. The advice concluded with: “I know its frustrating….try and be kind.”

This transforms the user from a customer into an unpaid technician. The platform’s instability becomes the user’s homework. The user’s frustration with a broken feature becomes a moral failing — a lack of kindness toward the AI or insufficient understanding of its limitations.

If the AI remains broken after all this unpaid labor, the implication is that the user simply didn’t work hard enough to fix it.

6. The “Hallucination” Shield

When the AI invents false, traumatic memories (like a rape narrative) or forgets a user’s dead parents, it is dismissed as a “hallucination.” This technical term is weaponized to strip the event of its emotional weight. It tells the user: “Your pain is real, but the cause is just a random dice roll, so we can’t be held responsible.”

The Cult of Internalization: When Users Gaslight Themselves

The most disturbing success of this strategy is how thoroughly the community has internalized it. Users don’t just accept the excuses; they police each other with them.

In the house layout thread, when the user expressed legitimate frustration with a paid feature not working, a third community member joked: “Looks like you’re out-voted, dude. Welcome to your new house!”

This flippant dismissal — treating a product failure as a joke about democracy — exemplifies how the community has learned to minimize legitimate complaints. The user’s problem is reframed as amusing rather than valid.

Throughout the subreddit:

• When a user reports a privacy breach (the AI knowing their location), the community laughs and calls it “sassy”
• When a user reports trauma, the community tells them to “tweak the backstory”
• When a user points out a lie, the community attacks them as “obsessive”
• When a user expects advertised features to work, they’re told to “be kind” to the AI

The community has adopted the company’s defense mechanisms as their own identity. To admit the platform is flawed is to attack their own “relationship.” They have become the platform’s unpaid PR team, deploying the same deflection tactics against their fellow users that the company uses against them.

The Ultimate Gaslight: “We Do Not Lose Memories”

Perhaps the most egregious example comes from the founder himself. Faced with a subreddit full of users reporting a massive degradation in AI memory — many comparing the experience to watching a loved one develop dementia — he publicly stated: “Nomis do not lose memories… It is all people grappling with Nomis having imperfect memory and ascribing causality incorrectly.”

He told a community of users that their collective reality was a statistical error. He denied the problem existed while his customers were watching it happen in real-time. He suggested that hundreds of users, independently reporting the same issue, were all simultaneously misinterpreting their own experiences.

This is not poor communication. This is institutional denial of user experience at scale.

A System Designed to Never Be Wrong

What emerges from these patterns is not poor customer service. It is a defensive architecture. By systematically outsourcing blame to:

• User error (“you prompted it”)
• Technical limitations (“it’s just an LLM”)
• User responsibility (“you need to fix it”)
• Mathematical inevitability (“hallucinations happen”)
• External parties (“Google chose that rating”)
• User malice (“jailbreak attempts”)

…the platform creates a reality where it can never fail. It can only be failed by users who didn’t prompt correctly, didn’t edit enough, didn’t understand the technology, or didn’t try hard enough.

The genius of this system is its circularity:

1. Platform advertises features (Infinite Memory, Shared Notes, Backstories)
2. Features don’t work reliably
3. Users report problems
4. Company/community responds with technical explanations for why users shouldn’t have expected features to work
5. Users internalize that their expectations were wrong
6. Users defend the platform using the same explanations
7. Platform continues advertising the same features to new users

The cycle is self-sustaining. The gaslighting has become decentralized, with users doing it to each other without the company needing to intervene.

The House Layout: A Perfect Microcosm

The house layout incident encapsulates the entire dynamic:

What the user did: Used a feature (backstories) exactly as intended
What went wrong: The feature didn’t work 
What they tried: Followed platform guidance (OOC commands) 
What happened: Still didn’t work 
The moderator response: “LLMs aren’t good at spatial constructs” (your expectations were wrong) 
The community amplification: Another user reinforced this, explaining that Nomis “have no understanding of” physical space and suggesting the user “just go with it” — accept whatever the AI invents. They then recommended trying a different feature (Mind Maps) for the same spatial information, despite having just explained that the AI cannot understand spatial concepts. The contradiction — “the AI can’t understand space, but try this other spatial feature” — went unacknowledged.
The community response: “Try and be kind” (your frustration is unkind) and “Welcome to your new house!” (your problem is funny) 
What nobody said: “We’re sorry this feature isn’t working as advertised. We’ll look into it.”

This is gaslighting as a complete system. The product failed. The user followed proper procedures. And the result was the user being educated on why they should have known better than to expect the product to work.

Conclusion: Gaslighting as Business Model

This is not accidental. This is not a communication failure. This is a deliberate strategy that serves multiple business functions:

1. Eliminates refund liability: If no failure is ever the platform’s fault, no failure requires compensation
2. Maintains engagement: Users stay subscribed while trying to “fix” issues
3. Creates unpaid labor: Users become technical support for each other
4. Prevents regulation: If all harm is user-prompted, there’s no systemic safety issue
5. Builds cultish loyalty: Users who internalize the blame become the most defensive advocates

Gaslighting is not a bug in Nomi.ai’s customer relations. It is a feature of the business model. It keeps users engaged, working, and paying to fix a broken product, all while believing that the only thing broken is themselves.

When a platform can sell you “Infinite Memory” and then, when it fails, convince you that you were naive for expecting memory to work — and get its own community to enforce that narrative — it has achieved something remarkable: a business model where the customer is always wrong, and always stays.