When your AI tool shuts down: a forced-migration playbook
Shutdowns in this category are frequent and fast. OpenAI discontinued Sora's web and app experiences on 26 April 2026 with its API ending 24 September 2026. When a vendor announces an end date, the first week decides how much you keep, and exports come before replacements.
By the AI Alt Finder team
A voluntary switch is a decision. A shutdown is a deadline someone else set, usually with a shorter notice period than you would like, and the sequence that works is the opposite of what people instinctively do. Export first, choose a replacement second.
How much notice do these actually come with?
Sometimes months, sometimes weeks. OpenAI published a two-stage timeline for Sora: the web and app experiences were discontinued on 26 April 2026, and the API is scheduled to stop on 24 September 2026, with account data deleted after those dates according to OpenAI's help centre notice.
Other closures have been sharper. After Meta acquired the PlayAI team in July 2025, the PlayHT platform shut down at the end of 2025, with user accounts, saved audio and voice clones deleted. Neither of these products should appear on any current shortlist, and both are useful precedents for how fast this happens.
Week one: get your data out
- Export everything immediately, before evaluating any replacement. Export windows close early and sometimes ahead of the announced date.
- Open the export and verify it. An archive you have not inspected is not a backup.
- Save the source material behind any trained asset. Cloned voices and trained styles are deleted with the account and cannot be exported.
- Download invoices and account records while the billing portal is still reachable.
Week two: choose a replacement on a shorter list than usual
A deadline is the wrong time for a thorough evaluation, so narrow deliberately. Two candidates, evaluated against the three tasks you actually perform, beats a comprehensive comparison you will not finish before the lights go out.
Weight longevity more heavily than you normally would. In a forced migration you are choosing where to land, and landing on another product with a short history means doing this again.
What signals actually predict a shutdown?
- Acquisition of the team rather than the product. Where a larger company hires the people, the product frequently follows them into retirement.
- A product that has been renamed repeatedly in a short period, which usually indicates unsettled positioning or external pressure.
- Silence: no releases, no changelog entries, no support responses over a quarter.
- Pricing changes that discourage new signups, such as a removed free tier with no replacement path.
- None of these are proof. All of them are reasons to keep your export current.
Building for the next one
The durable lesson is architectural. Where a tool sits behind an interface you control, replacing it is a configuration change. Where it is woven through a workflow with no abstraction, replacing it is a project.
For anything you depend on, keep a current export, keep the source material for trained assets, and know which single alternative you would move to. That preparation costs an hour a quarter and converts a crisis into an inconvenience.