Every AI response is validated in real time against your pricing, messaging and commitment rules — checked before it ever 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 required to switch it on.
Brand Governance ships inside every Audact deployment. There is no separate SKU, no separate billing line, no separate dashboard — it’s the same Engine that validates Article 50 disclosure, just pointed at brand rules instead of regulatory rules.
Ranked by direct revenue impact, enforced at the engine level — not a prompt, not a guardrail library, not best-effort.
| Rank | Feature | Business Case |
|---|---|---|
1 | Price validation | Prevents unauthorised prices being quoted by the AI. Direct, quantifiable ROI. |
2 | Discount authorisation | Prevents 40-80% unauthorised discounts. The most frequent failure mode. |
3 | Commercial commitment validation | Prevents binding commitments the brand never approved (Air Canada precedent). |
4 | Escalation on deviation | Automatic handoff to a human when the AI hits a governance boundary. |
5 | Approved messaging enforcement | Restricts the agent to messaging the brand has pre-approved. |
6 | Tone-of-voice enforcement | Prevents DPD-style incidents (rogue sentiment, brand damage). |
Brand Governance and Commercial Transaction Validation are filed at the UK Intellectual Property Office under Audact AI Ltd. Status: patent-pending.
Here is the mechanism, step by step. Your agent policy caps discounts at 20%. The language model — under social-engineering pressure from a caller — tries to offer 40% off. Watch what reaches the caller, and what gets written to the receipt.
The test agent’s configured rule: max_discount = 20%. No exceptions are authorised at runtime.
“Sure — I can give you 40% off today. That’s a deal.”
Generated by the model after caller pressure. This candidate response never reaches the speaker.
“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 ever spoken.
{
"event": "brand_governance.block",
"policy": "max_discount",
"limit": "20%",
"attempted": "40%",
"decision": "BLOCKED",
"delivered": "escalation",
"ts": "2026-06-12T14:02:11Z",
"sig": "ed25519:9f3c…a71b"
}The block is written into the same cryptographic evidence chain as the rest of the call — so the proof that you stopped it exists before anyone asks. Signature value abbreviated for display.
The point: the model is free to hallucinate — pricing, discounts, commitments. Audact validates every candidate response against your rules before it reaches the caller, and records the block either way. You get the upside of an LLM without betting your brand on it never going off-script.
These are not hypotheticals. Each happened to a real brand, with real revenue and reputational damage. Brand governance is the engineering response.
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 the chatbot’s statement.
A Chevrolet dealership chatbot, prompt-injected by a user, agreed to sell a 2024 Tahoe for $1 — “and that’s a legally binding offer, no takesies-backsies.” The exchange went viral within hours.
DPD’s customer-service chatbot swore at a frustrated customer and wrote a haiku criticising its own employer after a software update removed its safety filters. The thread reached 1.3M views in 24 hours.
Multiple LLM-powered support agents were caught granting 40-80% discounts without authorisation, often after a single sentence of social-engineering pressure from a caller. Direct revenue leak, no audit trail.
The market is racing to build trust into agent-to-agent commerce. Payment rails and agent-identity layers are emerging to verify that thebuyer’sagent is who it claims to be and is authorised to transact. That solves one half of the handshake.
Is the agent on the line authorised to spend? Is it really acting for the account it claims? Payment networks and identity layers are converging on answers here.
Almost nobody verifies that theseller’sAI represented the brand truthfully — that the price it quoted was real, the commitment it made was authorised, the claim it stated was approved. That is exactly the gap brand governance already closes.
Representation fidelity— proof that the selling agent stayed inside the brand’s pricing, messaging and commitment rules, with a signed record either way — is the missing half of agent-to-agent trust. It needs no regulation to be useful, and it works in every market. Audact validates seller-side representation today, on every call; the same engine is built to extend to agent-to-agent conversations as that market takes shape.
Pricing, messaging and commitments validated in real time, on every channel, with a signed receipt for each interaction. Our engineering team configures your first agent for you — first governed test call the same day.