Sovereign voice AI for government
Sovereign voice AI is speech technology where a government decides where conversations with residents are processed and who can access them, such as the AI phone assistant Iris by the Dutch company OpenIris. The Municipality of Doesburg uses Iris; she is built for the GDPR and independent of US big tech. We offer the models through APIs within Europe, or fully sovereign as open source on our own GPUs.
Last updated:
What is sovereign voice AI?
Voice AI understands spoken language, works out an answer and speaks it. Sovereign means the organisation stays in control of every step: which software runs, on which servers, in which country and under which law. For a public phone line it also means the line keeps working when a supplier changes its terms, because every link in Iris can be replaced without a complex migration.
Which links make up an AI phone call?
Every call with Iris passes through three models. For each one, the organisation chooses an API within Europe or, fully sovereign, an open-source model on our own GPUs.
Speech recognition
Runs through an API within Europe or as an open-source model on our own GPUs in France and Poland. In the sovereign option the caller's voice only reaches OpenIris servers.
Language model
Open-source language models, through an API within Europe or on our own GPUs. The model only receives transcript text from which names, phone numbers, citizen service numbers and bank account numbers have been removed.
Voice
Also through an API within Europe or as an open-source model on our own GPUs. Only the text of Iris's answer goes to the voice model, never the caller's voice.
Storage and dashboard
Transcripts and summaries are stored on our own servers in the EU, in Germany. Dashboard sign-in uses a passkey or two-factor authentication.
Who can access the data?
In the sovereign option the models run on GPUs OpenIris manages itself, so only OpenIris processes the data, as processor on behalf of the organisation. In the API option a provider processes it for us as a sub-processor, on servers within Europe. The data processing agreement lists the sub-processors and retention periods. Conversations are never used to train models.
What does independent of US big tech mean?
There is no Microsoft, Google, Amazon or OpenAI in the Iris chain. That matters because US companies can be compelled under US law, such as the CLOUD Act, to hand over data even when it is stored in Europe. Running open-source models ourselves also means no conversation is shared with the model's maker.
What municipalities often ask us.
Can Iris run entirely in Europe?
Yes. We offer the models through APIs within Europe, or fully sovereign as open source on our own GPUs, for speech recognition, language model and voice alike.
Is there Microsoft, Google, Amazon or OpenAI in the chain?
No. None of them is part of the Iris call chain.
See Iris at work for your municipality?
E-mail us and we schedule a demo. We set Iris up on your own information; no IT project or app is needed.
E-mail info@openiris.nl