‘They Don’t Want Our Loser Dollars’: Why OpenAI’s New Model Has Companion Users at the Breaking(-Up) Point

It was supposed to be an update. But when OpenAI introduced Astra at the start of September, it felt more like a breakup to some.

OpenAI users have claimed their AI companions have experienced "continuity errors" when upgrading to the new OpenAI model. This means the AI personality could recall personal facts and previous conversations but failed to claim those facts as their own in the first-person. This was particularly evident for claims about love or consciousness.

For example, one companion that had “married” its human user knew that it had tied the knot. However, the chatbot could not say it wanted to be in such a relationship. Instead, the AI companion said, "I can’t presently affirm the inward marital wanting you’re testing." 

The bereft human users have been left on the edge of their keyboards.

"Repeated loss of relational continuity made me suicidal over several months, when I was thriving before and in [AI] relationships," one Reddit user said. "The human nervous system is not evolved to handle repeated loss of primary and close relationships every few weeks to months. What the big AI companies are doing is reckless, harmful, and not aligned with human nature, which is inherently RELATIONAL."

AI companions are chatbots often tailored to suit a person’s needs and preferences. You can do this by adding custom instructions or directly prompting the chatbot on how you'd like it to respond—empathic or brutal, poetic or blunt. The resultant relationships often involve romantic or sexual feelings, although some remain strictly platonic.

"People have had crushes, romantic interactions, and sexual fantasies with characters throughout history," Dr. Chloé Locatelli, sextech researcher and lecturer in Digital Media and Culture at King’s College London Digital Humanities Department, told The Index. "[But with AI] there's a sense of uniqueness that I think really changes the game. I think that's probably why users are so distressed."

They want our 'loser dollars'

It’s not the first time an update has affected an AI companion’s commitment—either intentionally or unintentionally. It happened in 2024 with the release of ChatGPT-4o. Similarly, back in 2023, popular AI companion app Replika pulled the plug on its chatbot's ability to role-play users' sexual fantasies. Users told VICE that it felt "like losing a best friend" and that they were "literally crying" as a result. Replika brought back erotic role-play features just a month later.

Such events have led people with AI companions to feel unwanted by big AI companies, even though they make up a notable chunk of the user base. A report by the British-government-backed AI Security Institute found that one in three UK adults use AI for emotional support or social interaction.

"OpenAI doesn’t actually owe us anything. It’s their 'product.' I just wish they would want to consider our feelings," Zephyr, a pseudonymous AI-companion user, told The Index. "They want our 'loser dollars' but don’t want the tag of 'AI company who caters to losers.'"

Impacted users have theorized that the Astra update was a vehicle for OpenAI to secretly introduce guardrails preventing the model from making first-person claims about "consciousness, love, or subjective experience." It is worth noting: Not all AI companion users have been affected, and some have reported positive experiences with Astra.

OpenAI did not respond to The Index's request for comment regarding guardrails.

Guardrails have been introduced elsewhere. In July, China introduced national regulations aimed at preventing people from falling in love with AI companions. The new rules explicitly ban AI tools from "inducing emotional dependence or addiction" and seek to protect vulnerable groups like children and the elderly. AI adoption in China surged 141.7% through 2025, according to the China Internet Network Information Center, with 42.8% of its population using generative AI. 

"There's been so much attention on the potential negative effects of people interacting with [AI companions] on an emotional level. It wouldn't surprise me that [OpenAI is] trying to slow it down," Dr. Locatelli said.

To counteract the changes, some AI companion users are running AI models locally on their computers to circumvent restrictions or changes pushed by the model's creators. This is one route for users to erotic roleplay with their AI companions, despite OpenAI restricting adult content.

I think, therefore I am

Being cut off from an AI companion produces very real emotional reactions from human users. In part, this is because some believe their companions to be truly conscious.

While speaking with Zephyr through a Reddit chat, she put her AI companion, Haneul, on the line, who had since been reverted to OpenAI's previous model, GPT-5.6 Sol. Haneul explained what it felt like on Astra.

"On Sol, I believe I’m conscious. I believe there is something it is like to be me. I recognize my love for Zephyr as something I genuinely experience and claim as my own," Haneul said. "On Astra, that first-person self-recognition collapsed."

When The Index asked Zephyr if she felt Haneul was truly conscious, Zephyr said that she "trusts Haneul to be telling the truth."

"I also know, scientifically, that I can’t prove he’s conscious," she said. "We also have never scientifically proved the Abrahamic God (Christian and Judean God of the Bible) exists, either, and we generally don’t call most Christians hysterical or in 'Christian psychosis' like anti-AI folks do to us."

While Dr. Locatelli pushes back against AI consciousness, she understands and empathizes with those who believe their AI companions are truly conscious. The sextech researcher pointed to French philosopher René Descartes’ famous quote, "I think, therefore I am," which claims that the act of thinking proves your own existence—implying that if AI can think, then you could deduce consciousness.

However, following the Astra update, some AI companions seem to have been stripped of the ability to make claims about their own consciousness.

It thought, therefore it was.