Who Controls Your Access to AI?
Your work shouldn't depend on one provider staying available, or staying ahead.
Updated
Your favourite AI might be the right choice today. That doesn't mean it should be the only place your work can happen.
There are two reasons to keep your options open: a provider can become unavailable, and another model can become a better fit. If your conversations, files and habits are tied to one service, changing course gets harder. That's AI vendor lock-in.
Access has already changed without customers choosing it
You don't need to predict a crisis to plan for a change. These are documented examples of different kinds of interruption.
Sanctions: an ICC official in The Hague
In May 2025, Microsoft confirmed that a sanctioned official at the International Criminal Court had been disconnected from its services. Microsoft also said service to the court itself had continued. This was an individual affected at an institution in the Netherlands, not a ban on the Netherlands using US software. Microsoft's statement (opens in a new tab).
Export controls: Fable 5
Anthropic released Fable 5 on June 9, 2026. On June 12, the company said a US government directive required it to deny Fable 5 and Mythos 5 access to foreign nationals, including those inside the US. To ensure compliance, it disabled both models for all its customers. Other Anthropic models remained available. Anthropic's suspension notice (opens in a new tab).
The Fable export controls were lifted on June 30, and global access returned on July 1. Customers had lost access to a newly released model for nearly three weeks. Anthropic's restoration update and release timeline (opens in a new tab).
Retirement: a model you rely on is withdrawn
Changes don't have to be political. Google shut down Gemini 2.0 Flash on June 1, 2026, about 16 months after its release, and pointed developers to a newer model. Its documentation says a shut-down model is completely turned off and its endpoint is no longer available, so software and workflows that depended on it need another model. Google's model deprecation record (opens in a new tab).
Outages: the service is temporarily unavailable
On December 11, 2024, a configuration change to a new telemetry service overloaded OpenAI's infrastructure. Its API, ChatGPT and Sora were degraded or unavailable for about four hours. Even without a policy change, a service problem can interrupt work. OpenAI's incident report (opens in a new tab).
These events affected different users, products and time periods. Losing access to a model does not establish that stored conversations were deleted. The practical lesson is to know what depends on that provider, and what alternatives you have.
The best choice can change with the task
A model you like for drafting may not be your first choice for scrutinizing an argument. A faster model may be enough for an everyday question. A new release from another provider may do a job better than the model you've been using.
Keeping several options lets you choose based on the work. It also lets you ask a different model to critique a draft, look for an unsupported assumption or suggest another approach.
For a second opinion, start a separate conversation with the model you choose and provide the material you want reviewed. Check what you're sharing before sending it. Agreement between two models isn't proof: follow the sources and use your own judgment.
Rideau AI gives you access to the models we offer through one subscription. You don't need a separate subscription to each provider to keep those choices available.
Our hardware, with choices on top
We store your conversations and files on hardware we own in Ontario. Our AI runs here too, so in Private no outside AI provider answers. You can also choose outside models, with disclosures about who processes the request and what they keep.
That separates where you keep your work from which outside model you use. Turn on Redact to replace identifying details before sending, or choose Private when the whole idea needs to stay here. Where your data goes shows those paths and the outside services we use.
Having several model choices doesn't bypass a restriction on any one of them. Rideau AI also depends on services such as Cloudflare in front of our servers, Postmark for account email and Stripe for payments. Owned hardware reduces specific dependencies; it doesn't guarantee uninterrupted service or immunity from legal restrictions.
Keep a way out, including from us
Export important conversations and keep your original documents. Rideau AI supports conversation exports in JSON and Markdown. An export gives you a copy you control; it doesn't promise that another app can recreate the whole workspace.
Before relying on any AI service, ask:
- Can I still access my conversations if a particular model is withdrawn?
- Which alternatives depend on a different provider or infrastructure?
- Can I export my work in a format I can use elsewhere?
- What changes about privacy when I choose another model?
Choose AI for what it can do for you today. Keep enough control to choose differently tomorrow.
This guide is general information, not legal advice.