AI Use and Safety Policy

AI may appear in customer chat, phone workflows, setup help, docs assistant, plan advisor, summaries, routing and support tooling.

Different AI surfaces have different roles. The public helper answers product questions; the customer receptionist answers business-specific customer questions where configured.

AI should answer from configured business knowledge, docs, plan data and safe product context. If context is missing, the correct behaviour is to ask for clarification or escalate.

Customers must keep services, prices, policies, opening hours and escalation contacts accurate.

Customer-facing AI phone workflows must include the AI and recording disclosures required in your jurisdiction. Businesses must not present an AI assistant as a human agent.

Users should know when they are using an assistant and when a human will review or take over.

Complaints, safety issues, emergencies, payment-card details, legal/regulatory wording, medical topics, chargebacks, customer distress and unclear answers should be handed to humans.

SIMCOAI does not replace professional advice, regulated decision-making or your own customer duty of care.

Where SIMCOAI answers a phone call on a business’s behalf, the caller may ask for a person at any point in the conversation, and the AI will route the call to a human, a callback queue or the business’s own escalation contact according to how that business has configured escalation. The caller does not need to justify the request, exhaust the automated flow first, or be refused twice before it is honoured.

This is a standing right the caller can exercise unprompted, not a fallback that only fires when the AI decides it is stuck. It applies in addition to, not instead of, the disclosure required in section 03 and the escalation triggers a business configures under section 04.

On phone-enabled plans, an authorised team member can select Take over by voice in Live Calls and use their browser microphone and speakers. The caller stays with the receptionist while the browser connects. Once the voice handoff connects, AI speech recognition, transcription and replies stop for the human conversation. An escalation record or callback request alone does not mean a team member has joined. Availability still depends on your team and configured escalation route.

AI outputs are probabilistic and can be wrong, incomplete, out of date or unsuitable for a particular case. To the maximum extent permitted by law, SIMCOAI accepts no liability for any error, omission, inaccuracy, decision, action or outcome produced or actioned by the AI — including refunds, bookings, cancellations, escalations, order updates, quotes, payments, transcriptions or advice. The AI is provided “as is” and “as available” with no warranty of accuracy or fitness for purpose.

You are responsible for supervising, configuring and reviewing AI behaviour, keeping human fallback and approval controls in place, and validating any output before it is relied on or takes external effect. This policy forms part of, and is subject to, the disclaimers and limitation of liability in our Terms of Service. Nothing here excludes liability that cannot lawfully be excluded (such as death or personal injury caused by negligence, or fraud).

Test realistic questions before launch, review logs after launch, remove repetitive scripts, shorten robotic wording and update knowledge when answers are vague.

AI can make mistakes. Do not advertise guaranteed accuracy, guaranteed compliance, guaranteed revenue or full replacement of staff judgement.

We test AI responses against realistic questions before a new business goes live, and periodically review real conversation logs for tone, accuracy and consistency afterwards. Where a business’s own configuration, or a change made by a managed AI provider, produces systematically unhelpful, evasive, or unequal treatment of similar requests, that is treated as a quality defect to be corrected — not an acceptable trade-off of an automated system.

Do not use SIMCOAI to produce deception, impersonation, spam, unlawful surveillance, discriminatory decisions, unsafe instructions, credential capture or regulated advice without written review.

Do not use prompts to bypass safety systems or expose private information.

SIMCOAI Cura 1, Sermo 1, Mens 1 and Vis 1 are names for SIMCOAI-managed assistant configurations. SIMCOAI Viva 1 is an experimental speech-to-speech configuration and is not available in the standard model selector. These names do not mean SIMCOAI trained the underlying base models. Model choice does not change the human review, disclosure or prohibited-use requirements in this policy. See model descriptions and the privacy policy for processing disclosures.

SIMCOAI may use AI providers such as OpenAI. Provider availability, model behaviour, safety systems and rate limits can affect answers and latency.

Where providers are degraded, SIMCOAI may fall back, delay, block or route to support.

SIMCOAI does not use your business data, customer conversations, call recordings, transcripts or uploaded evidence to train, fine-tune or otherwise improve its own AI models. SIMCOAI does not operate any home-grown or self-trained model — the AI features described in this policy are provided through OpenAI’s hosted API under SIMCOAI’s own commercial agreement, using SIMCOAI’s own managed credentials as set out in section 08.

SIMCOAI has not enabled, and does not intend to enable, any model-training or data-sharing option offered by its managed AI provider. Where a provider’s own published policy states that API requests are not used to train its models by default, SIMCOAI relies on that commitment rather than a bespoke arrangement, and has taken no separate action that would change it.

This section describes SIMCOAI’s own conduct and its provider’s standard terms; it does not amend or replace the provider’s own published data usage policy, which governs directly and may be updated by that provider independently of SIMCOAI. If SIMCOAI’s own position changes, we will update this section and, where the change is material, treat it as a policy change under section 11.

Where a business enables proof uploads, SIMCOAI may read customer-submitted evidence — photographs, receipts and documents — and compare a barcode, tracking or reference value against the value the business said it expected. Results are shown as matched or unmatched alongside an analysis status.

These checks are assistive and are not a determination. They exist to help a human reviewer, not to replace one. Accuracy is affected by image quality, lighting, damaged or partial barcodes, similar reference formats and document variation, so both false matches and missed matches are possible. A match is not evidence of authenticity, ownership, condition, delivery or entitlement, and an unmatched result is not evidence of wrongdoing.

No automated final decisions. SIMCOAI does not approve, deny or issue refunds, and does not confirm or cancel bookings or close escalations on its own. Sensitive workflow actions are protected by verification codes and the platform only offers the actions currently available on a record; a person must review the evidence and approve the outcome. The result is governed by the business's own refund, returns and booking policy and by applicable consumer law.

SIMCOAI checks must not be described to end customers as certification, identity verification, fraud detection or independent proof, and must not be used as the sole basis for accusing a customer of dishonesty. Businesses remain responsible for how they communicate outcomes, and the human review commitments in section 04 apply in full to evidence-assisted workflows.

AI configuration, models, prompts and safety wording may change to improve reliability, speed, security and compliance posture. Material policy changes may require renewed acceptance.

A managed AI provider updating the underlying model, or changing its own safety behaviour, is not by itself a change to this policy. Where that happens, we may switch the model in use, adjust prompts and safety wording to compensate, or notify affected businesses — without that alone requiring renewed acceptance. Renewed acceptance is reserved for a change to what SIMCOAI itself commits to in this document.

The assistant uses the Customer’s saved knowledge, workflow settings and conversation context to produce a response. Those sources can be incomplete, outdated or inconsistent. The Customer should keep them current and define what the assistant should do when a fact is missing. SIMCOAI may add technical safeguards that prefer a clarification, handoff or refusal over an unsupported assertion. No configuration can guarantee that every generated sentence is correct; a human should review a consequential output before acting on it.

The assistant should distinguish a proposed action from a completed one. If a booking, refund, order or message requires an external confirmation, the record of that confirmation is the reliable evidence of completion. A fluent answer alone is not proof that a system of record changed. Customers should train staff to inspect the record when a caller disputes an outcome.

The Customer is responsible for deciding what business information may be given to the assistant and what may be repeated to callers. It should avoid uploading secrets, another person’s confidential material or data it has no right to use. SIMCOAI may process that material to provide the requested service under the agreement; submitting it does not transfer ownership to SIMCOAI. We may refuse material that appears unlawful, malicious or likely to expose another customer’s information.

Where a source document changes, the Customer should update or remove the old version and check the assistant’s answer to a representative question. It should not assume that a new upload automatically resolves a contradiction with older instructions. A support report about an incorrect answer should include the relevant source and expected correction, without sending more personal data than needed.

A qualified person should review an AI suggestion before it determines a refund, rejects a complaint, changes a sensitive account detail, gives professional advice or affects a person’s significant interests. The review should include the underlying facts, applicable business rules and any evidence the person submitted. A reviewer should be able to change or reject the suggestion and record the reason. An apparent pattern of errors should lead the Customer to pause or narrow the workflow while it is investigated.

SIMCOAI may provide a handoff, approval or audit control to support this process, but the Customer decides how to staff and use it. We will not claim that a tool’s availability alone makes the Customer’s whole decision process lawful. If a human review is legally required, the Customer must make it meaningful rather than simply approving every suggestion.

We may use limited examples, aggregate measures and error reports to test reliability, safety and user experience as described in the Privacy Policy and data processing terms. Access to a Customer’s identifiable content for support or evaluation should be limited to a legitimate need and protected by appropriate controls. We should not present a test result as a guarantee of performance in every accent, language, business or network condition. A Customer may use available feedback and logs to identify a recurring problem.

When we identify a material safety or correctness issue, we may adjust a prompt, disable an affected capability or require an approval step while we correct it. We will explain a material customer-facing restriction through appropriate product or support channels. Corrections can reduce risk but cannot replace accurate Customer knowledge and appropriate human oversight.

A person should be able to understand when they are dealing with an automated assistant and how to reach a human for a matter the assistant cannot safely complete. The Customer is responsible for its own notices about the purpose of collection, recording and use of personal data. SIMCOAI may provide default wording or controls, but the Customer should adapt them to its activity and applicable law. The assistant should not falsely state that a human has reviewed or approved an action.

If the interaction changes from AI to a live person, the business should manage the handoff so context is available only to those who need it. A caller’s request for a person should be respected through the configured path where feasible. An automated channel must not be used to trap someone who needs to complain, correct data or exercise a legal right.

Do not instruct the assistant to discriminate, deceive, harass, collect unnecessary sensitive data, impersonate a person or conceal uncertainty about an important fact. Do not use it as the only decision maker for a legally significant outcome without a separate lawful basis and suitable safeguards. We may reject an instruction or output that appears to create a serious safety or legal risk, even if the Customer has entered it in its own configuration. We should explain the restriction when doing so would not undermine security or abuse prevention.

The Customer should assess a new use case before launch, including foreseeable failure modes, affected people and a fallback path. A use that is technically possible is not automatically approved by SIMCOAI or compliant with the Customer’s obligations. The Acceptable Use Policy and Terms govern enforcement and remedies.

People may use accents, mixed languages, noisy connections, unclear dates or unfamiliar names. The assistant may misunderstand a detail even when the transcript looks plausible. A Customer should test representative language and accessibility needs in its own audience, offer clarification, and keep a human route for a person who cannot use the automated one. The assistant should not guess a booking date, price or identity from a weak signal when the consequence matters. A caller should be able to correct a misunderstood fact without restarting an entire complaint.

SIMCOAI may improve speech and text handling, but published model or voice names do not promise identical performance across every language or device. We should explain a known material limitation and support a safer fallback.

An AI-generated sentence saying a task is done must reflect a confirmed system event. If the external calendar, payment provider or business system did not confirm the action, the assistant should present it as pending or unresolved. The Customer should not ask it to conceal an error or imply a human checked something that no human checked. We may keep a limited event record to help reconcile an ambiguous result. A Customer should review that record before repeating an action that could create a duplicate charge, booking or message. The obligation to be accurate about completion protects both the business and the person relying on it.

A Customer should repeat representative tests when it changes prices, services, staff rules, escalation destinations or business knowledge. We should test material platform changes before release and monitor reported failures. A test should include an ordinary request, an ambiguous request and a request the assistant must hand off. The Customer should verify the resulting records, not only the spoken or written answer. If a change causes repeated wrong outcomes, the affected workflow should be narrowed or paused while corrected. We may provide an update about a known material issue and a safer configuration. No single scripted test can establish that all real conversations will succeed.

If a person disputes an AI answer, the Customer should check the underlying conversation, saved business rule and confirmed transaction rather than relying on a summary alone. It should correct a wrong record, tell the person the accurate outcome and consider whether others may have received the same error. SIMCOAI will investigate a reproducible platform issue and may request a limited example. We should not require the Customer to disclose unrelated conversations to show one fault. A correction to future answers does not by itself reverse an earlier charge, booking or decision; the relevant system of record and refund or complaint process must handle that consequence. Both parties should keep a clear record.