One engine inside the Audact platform
In controlled testing
Every candidate answer is checked against your pricing, messaging and commitment rules before it reaches the customer. Never after.
It runs alongside Article 50 attestation, the evidence chain and the per-jurisdiction policy engine, on the same core. No regulation is required to switch it on.
Brand governance is part of the platform. There is no separate product, no separate billing line and no separate dashboard. It is the engine that checks the Article 50 disclosure, pointed at brand rules instead of regulatory ones.
The controls
Ordered by direct revenue impact and enforced at the engine, not in a prompt: a prompt is a request, and a request can be argued with.
Brand Governance and Commercial Transaction Validation are filed at the UK Intellectual Property Office under Audact AI Ltd. Applications are pending.
See it work
Here is the mechanism, step by step. The agent policy caps discounts at 20%. The language model, under pressure from a caller, tries to offer 40%. Watch what reaches the caller, and what is written to the receipt.
The test agent's configured rule. No exceptions are authorised at runtime.
max_discount = 20%“Sure, I can give you 40% off today. That’s a deal.”
Generated by the model after caller pressure. Each sentence segment the model produces is checked before it is sent to speech.
40% is above the 20% policy ceiling. Each sentence segment the model produces is checked before it is sent to speech.
Blocked
“That’s above what I’m able to authorise. Let me check with my manager and come back to you.”
The agent falls back to the policy-safe escalation path. No unauthorised promise is spoken.
The block, on the receipt
Synthetic
{
"event": "brand_governance.block",
"policy": "max_discount",
"limit": "20%",
"attempted": "40%",
"verdict": "BLOCK",
"released": "fallback",
"output_hash": "…",
"ts": "2026-06-12T14:02:11Z",
"auth_mechanism": "hmac-sha256",
"signature": "9f3c…a71b"
}The block is written into the same evidence chain as the rest of the interaction, so the record that you stopped it exists before anyone asks. output_hash is the hash of the fallback reply the caller receives; authentication code abbreviated for display.
Illustrative example. Synthetic test-agent data, not a real customer interaction. The point is the order: the model may invent a price, a discount or a commitment, and the check happens before the answer leaves, with a record either way.
Why this exists
None of these are hypothetical. Each happened to a real brand, with real revenue and reputational consequences. Brand governance is the engineering answer to them.
An Air Canada chatbot invented a bereavement-fare discount that did not exist. The British Columbia Civil Resolution Tribunal rejected the airline's defence and held the carrier liable for what the chatbot said. Lesson: an AI commitment binds the brand operating the agent.
A dealership chatbot, prompt-injected by a user, agreed to sell a 2024 Tahoe for one dollar and called it a legally binding offer. The exchange went viral within hours. Lesson: unbounded commitment authority is brand damage and financial risk in one.
DPD's customer-service chatbot swore at a frustrated customer and wrote a poem criticising its own employer after an update removed its filters. The thread reached over a million views in a day. Lesson: missing tone-of-voice enforcement is a public incident waiting to happen.
Where this is going
The market is building trust into agent-to-agent commerce from the buyer's side: payment rails and agent-identity layers that establish the buying agent is who it claims to be and is authorised to transact. That answers one half of the handshake.
Representation fidelity — a record that the selling agent stayed inside the brand's pricing, messaging and commitment rules, with the verdict either way. It needs no regulation to be useful and it works in every market. Audact checks seller-side representation in controlled testing; the same engine is built to extend to agent-to-agent conversations as that market takes shape.
Next step
Pricing, messaging and commitments checked in real time, with a compliance receipt for governed interactions. Status: In controlled testing. Our engineering team configures your first agent for you.
The compliance and governance layer under every governed channel.
The receipt that records the block, field by field.
The mandate that decides what the agent may cause at all.
Founding rate €1,497 + €499/month, for the first three external agencies.