Cancel AI subscription: why users cite cost, privacy, regressions

Simon Willison published a May 31 post describing why he cancelled his paid AI subscription after months of trial. He framed cancellation as a simple fix for immediate problems: rising cost, waning quality from model updates, and unclear data-handling practices.

Willison’s case is framed as a practical choice rather than a principled boycott. He documented that removing the subscription reduced friction in his day to day work, eliminated recurring charges, and avoided the need to chase changing product behavior. The post joined a string of public accounts from independent developers and power users describing similar decisions.

What users are finding

Three recurring threads appear in user accounts that lead to cancellation. First, cost. Subscription pricing combined with token or usage-based metering has left some users facing unpredictable bills for tasks that previously felt inexpensive. For people using models for routine scripting, research or drafting, small per-call increases add up quickly.

Second, performance regressions. Several users report that model updates intended to improve capability sometimes change how models respond in ways that break existing workflows. Prompts and shortcuts that worked reliably can stop producing useful output after an update, forcing users to rebuild prompt templates or revert to older patterns.

Third, privacy and data handling. Unclear or evolving terms around telemetry, training data retention and commercial reuse have made some customers uncomfortable sharing sensitive code, notes or drafts with hosted services. For those handling client work, even a low probability of exposure can be enough to cancel.

Other motivators include product changes that remove favored features, longer latency when models are throttled, and a growing preference for single-purpose tools that solve narrow tasks better than a general assistant.

Alternatives and next steps users choose

Cancelling does not always mean abandoning AI assistance. Many users downgrade to free tiers for casual queries while shifting heavy work to a mix of approaches. Common alternatives include using open-source models hosted locally or on low-cost cloud instances, building and maintaining prompt libraries that work with smaller models, or adopting focused tools for tasks such as code completion or document summarization.

For some teams, the calculus leads to hybrid setups: keep a minimal paid plan for high-capacity needs, move repeatable tasks to self-hosted models, and reserve the paid service for occasional large-context operations. Others choose to wait for clearer guarantees on pricing and data use before re-subscribing.

Vendors are taking note. Product forums, changelogs and policy updates have become focal points for users deciding whether to stay. Where vendors provide predictable pricing, model versioning options and explicit data controls, churn appears lower. Conversely, opaque changes to billing or model behavior are frequently cited in exit posts.

Why it matters

Rising cancellations shift the leverage back toward users who demand stable pricing, version controls and clearer data guarantees. The trend will especially affect freelancers, small businesses and developers who cannot absorb unpredictable costs or frequent workflow breakage. Vendors that want long-term paid customers will need to offer clearer guarantees and simpler migration paths between plans and model versions.