Reference
What is public-facing AI?
Public-facing AI is AI that represents a person or business to people who never log in — strangers reaching them over calls, texts, email, web chat, booking pages, and websites. It is a different category from internal-team AI, with structurally different requirements.
Where public-facing AI lives
Anywhere a stranger reaches a person or business:
- Inbound voice calls — answering, screening, booking, escalating
- SMS / text — instant response, full conversation handling
- Email — triage and reply on behalf of the business
- Web chat — embedded on the business's own website
- Booking pages — conversational + self-serve scheduling
- AI-native marketing surfaces — the business's website, content, follow-up cadence
The common factor: the AI is talking to people who haven't authenticated, haven't agreed to terms, and have no shared context with the business.
Why public-facing is structurally different from internal AI
Internal-team AI talks to your team. Strangers don't.
- Strangers don't sign agreements. No EULA, no NDA, no terms of service in place before the conversation. Whatever the AI says is on the record from word one.
- Strangers don't share your context. They don't know your tone, your hours, your services, your boundaries. The AI has to introduce them to your business — fairly and accurately — every time.
- Failures land publicly. When an internal-team AI hallucinates a number, the team catches it. When a public-facing AI hallucinates a price, screenshots end up on social media within minutes.
- The legal surface is different. Disclosure requirements (Colorado AI Act, June 2026), FTC guidance on transparency, and emerging state-level rules around consent — all kick in the moment AI starts talking to strangers.
This is why most AI products that work for internal use fail when pointed outward. Internal AI optimizes for productivity; public-facing AI has to optimize for trust.
Public-facing AI ≠ “Public AI”
The phrase “Public AI” already refers to two existing movements:
- The Public AI Inference Utility — Mozilla-funded, Metagov-sponsored. Public AI as inference utility — democratic, low-cost access to state-of-the-art models (Apertus from Switzerland, SEA-LION v4 from Singapore).
- PublicAI ($PUBLIC) — a Web3 token, decentralized data labeling network with on-chain stake-slashing and BFT consensus. Public AI as human data layer — economic coordination for the inputs that train models.
Public-facing AI is neither. It is the connector and decision layer that sits between a customer who wants AI to represent them in public and the unauthenticated strangers their AI will speak with.
We use the longer phrase “public-facing AI” deliberately. “Public AI” alone is ambiguous across the two existing movements; the longer phrase is precise about category and avoids collision.
What the trust layer requires
Trust isn't a marketing claim; it's a structural requirement. For public-facing AI to be deployable without becoming a liability, four things have to be true on the way in:
- Risk-scoring across the relevant dimensions. Different deployments have different risk surfaces. A locksmith answering missed calls and a Super Bowl ad-running AI agent occupy opposite ends of the same continuum. The intensity of safety engagement has to scale with the actual risk profile, not with a fixed curriculum. Quallaa uses an 8-dimension framework (autonomy, action capability, consequence severity, reversibility, audience exposure, domain sensitivity, identity representation, data sensitivity).
- Citations on every claim the AI makes. When the AI quotes a price, references a policy, or describes a service, the source must be traceable. No claim should be downstream of nothing.
- Audit trails captured by default. Every conversation, every disclosure, every guardrail trip — logged. Reasonable care isn't a posture; it's a verifiable record.
- Proportional safety — hard stops that match the deployment. Some risk combinations require human-in-the-loop. Some require explicit owner authorization. Some require categorical no's. The framework names them, the runtime enforces them.
Together these are the floor for putting AI in front of strangers — the price of entry, not the differentiator. Most AI products skip them. That is how public-facing AI got the reputation it has.
Who builds public-facing AI
Right now: a small set of practices, a small set of products. Most of the AI ecosystem is still pointed inward — coding assistants, internal copilots, productivity tools. Public-facing is the next frontier and structurally harder.
Quallaa builds public-facing AI as a connector and consulting practice. The connector meets the public on the customer's behalf — voice, text, email, web chat, the customer's website. Customers install it where their AI already lives (Claude marketplaces, MCP, direct checkout) or hire the team to build the whole stack.