Our approach to ethical AI in AAC
At Flexspeak, we believe that Artificial Intelligence (AI) has the potential to make AAC faster, more personal, and more accessible: especially for multilingual communicators and families who have long been underserved by existing tools. Used well, AI can reduce setup time, support AAC users and stakeholders who communicate across languages, and help clinicians scale their expertise across larger caseloads.
But these benefits must never come at the cost of a user's voice, agency, or identity.
We also believe AAC users deserve the same agency and autonomy as any other communicator. Anyone else can choose whether to use AI in their writing, translation, or daily workflow. AAC users should have that same choice - to use AI, to not use it, or to use it selectively. Our job is to make that choice real, not to make it for them.
We believe AI should support people who use AAC: not speak for them, not guess at their meaning, and not override their intent. We are committed to using AI in a way that respects the rights, dignity, and individuality of every person who uses our technology.
Our principles
The following principles guide how we develop and use every AI feature across Flexspeak AAC and Flexspeak Studio, today and in the future:
1.
AI should assist, not replace, human communication. The user is always the communicator. AI can help draft, translate, or customize: but the words spoken belong to the person using the app.
2.
The user is always in control. Flexspeak is designed so that the user shapes every AI interaction. In features like AI Assist, the user selects the communication function (for example, making a request, commenting, or protesting) before any message is generated - so AI follows the user's intent rather than guessing at it. Whenever AI generates new content, the user reviews and validates that content before it is saved, applied, or spoken.
Some automatic transformations happen in the background to make AAC practical - for example, when saving a tile, grammatical conjugations like "fetched" or "fetching" are generated from a base word the user has entered (such as "fetch"), and translations of user-approved content are generated across the languages a user has enabled. These transformations are always derived from content the user has already chosen.
3.
AI features are optional. Flexspeak's essential AAC features work fully without AI. Users can communicate by pressing tiles, building messages, and navigating boards using non-AI voices and tools. AI-powered features can be turned on or off at any time.
4.
Built with the people who use it. AI features in Flexspeak are designed and reviewed by a team that includes licensed speech-language pathologists. But the most important voices in shaping how we use AI are AAC users themselves. We actively seek feedback from the AAC community and refine our features based on what we hear.
5.
Transparency about where AI is used. We are specific about which features are AI-powered so you always know when AI is involved. A current list of AI features is maintained in our AI Terms, and we surface AI use in-app where features are introduced.
6.
Your data stays yours. We maintain signed Business Associate Agreements (BAAs) with every AI provider we work with, including OpenAI, Google, and Anthropic. Under these agreements, your communications are never used to train AI models and are never shared for purposes beyond delivering the service. Depending on the provider, data is processed under a Zero Data Retention workflow or retained only briefly for automated safety and abuse monitoring, after which it is automatically deleted. No data is used for model improvement or made available for human review outside of automated safety checks. You own your content. We are aligned with FERPA, COPPA, and IDEA.
How we use AI
You will encounter AI in Flexspeak across several categories, including:
Real-time translation: helping AAC users and their stakeholders communicate across languages
Content generation for customization (including, but not limited to: Quick Add, Quick Folders, Vocabulary Builder): helping AAC users, clinicians, and families build boards, tiles, and other content faster.
Predictive support (including, but not limited to: AI Assist): helping turn partial selections into complete messages.
In every category, the user reviews and validates AI output before it is saved, applied, or spoken.
Ongoing work
We know this is an evolving space, and we know the AAC community has well-founded concerns about how AI is being introduced into communication tools. We take those concerns seriously, and we believe AAC users should have a voice in shaping how AI shows up in the tools they rely on.