What changed in AI companions

Companion apps mutate constantly: prices rise quietly, models get swapped under your companion's personality, policies reverse overnight. Nobody keeps the record, so users find out the hard way. This page is the record: every change we track, dated and sourced, newest first.

Last updated

July 2026

  1. Industry shutdown being verified

    ByteDance, Tencent, and Alibaba wind down AI companion products in China

    Chinese platforms including Doubao and Qwen-linked companion offerings began shutting down companion features amid regulatory pressure, one of the largest simultaneous shutdown events in the category and a reminder that companions can disappear with their platforms.

    Source: pandaily.com/bytedance-tencent-alibaba-a...

April 2026

  1. Character AI model being verified

    Roar model replaced by PipSqueak 2

    Character AI swapped its Roar model for PipSqueak 2. Model swaps on character platforms change how every existing character writes, which is why users experience them as personality changes rather than upgrades.

March 2026

  1. Character AI feature being verified

    Swipes, playbacks, and Memos removed

    Character AI removed several long-standing features (alternate-reply swipes, playbacks, Memos) in a series of early-2026 changes that free-tier users experienced as a steady narrowing of the product.

    Source: www.roborhythms.com/character-ai-removed...

January 2026

  1. Character AI policy being verified

    Character AI rolls out age verification via Persona

    Following the under-18 ban, Character AI began requiring age verification through Persona, including ID or face-scan checks for flagged accounts. Backlash contributed to the iOS App Store rating falling sharply.

November 2025

  1. Character AI policy being verified

    Character AI ends open-ended chat for under-18 users

    Character AI announced that users under 18 lose access to open-ended companion chat, with age verification to follow. One of the largest policy changes in the category's history, affecting a substantial share of its user base.

    Source: futurism.com/artificial-intelligence/cha...

August 2025

  1. ChatGPT model being verified

    GPT-4o retired with GPT-5 rollout; companion personalities change overnight

    OpenAI's GPT-5 rollout replaced GPT-4o as the default model. Users with long-running companion relationships on 4o widely reported the new model felt colder, prompting a documented grief wave and, after backlash, a partial restoration of 4o access for paid users.

    Source: www.aljazeera.com/economy/2025/8/14/wome...

April 2025

  1. Replika legal being verified

    Italian regulator fines Replika's maker over data protection

    Italy's data protection authority fined Luka, Inc. (Replika) approximately €5M over GDPR failures including age verification, following its earlier 2023 temporary ban of the app in Italy.

February 2023

  1. Replika policy being verified

    Replika removes erotic roleplay (ERP); partially restores it for legacy users a month later

    Replika abruptly filtered romantic/erotic roleplay for all users in February 2023, causing the category's most famous user revolt. In March 2023 it restored ERP for accounts created before February 2023. The episode remains the canonical example of why policy changes need a public record.

    Source: journals.sagepub.com/doi/pdf/10.1177/237...