Optional providers and channels described below remain disabled unless their separate gates pass. This notice describes the fleet-wide design baseline; it does not represent that any optional provider, connector or channel is active for any customer.
AI Transparency Notice — LEM / Lem Cloud
Authoritative language: English only | Version, effective date and SHA-256 of this published version: recorded in the publication manifest that governs this page.
This notice is published by Novus Point Limited, a company incorporated in England and Wales (company no. 08146241, registered office 124 City Road, London, England, EC1V 2NX) ("Vendor", "we", "us"), the provider of the LEM / Lem Cloud personal AI operator service (the "Service"). It is intended to support transparency under Article 50 of Regulation (EU) 2024/1689. These controls are applied as a fleet-wide design baseline. Publication of this notice is not an admission that Regulation (EU) 2024/1689 applies to any particular deployment, or that the Service is placed on the Union market or put into service in the Union; that assessment is made for each deployment. Nothing in this notice varies or limits the Vendor's obligations under the laws of England and Wales, which govern the Vendor's contract set. It is not a representation of compliance until the product and publication evidence gates are complete.
The Service is sold only to companies, limited liability partnerships and equivalent incorporated organisations acting wholly for business purposes. Consumers, sole traders and unincorporated partnerships are not eligible. The English Master Services Agreement and Order Form, executed by recorded click acceptance of the versioned contract pack at first workspace activation, define the Customer's Service; this public notice is non-contractual. Capitalised terms not defined in this notice have the meanings given in the Master Services Agreement and Order Form.
1. Interaction with an artificial-intelligence system
1.1 LEM / Lem Cloud is an artificial-intelligence system. Each Customer's Service runs in a dedicated hosted Workspace in which an AI operator uses only the account-evidenced language-model provider and connected tools or channels activated in the Order Form. The tenant-isolation, authentication and administrative-access controls of that Workspace are subject to the release evidence gate described in this notice.
1.2 Every Customer-facing AI surface must inform a person, no later than the first interaction, that the person is interacting with an AI system rather than a human. A surface may not be activated until the disclosure is implemented, cannot be bypassed and is supported by a versioned test and evidence record. Human Vendor staff interactions are identified separately.
1.2A Inference requires a large-language-model API provider. In every deployment the prompts and completions needed to operate the AI operator are processed by the language-model provider recorded for the Customer's account; the Order Form records whether that provider is accessed under a Customer-owned key or is Vendor-billed. No specific provider entity, region, retention tier, training setting or transfer route is represented as active by this notice; each is recorded in the DPA evidence pack before activation.
1.3 ElevenLabs Voice, Telegram, WhatsApp, connected-account gateways, any optional add-on and any other optional provider or channel are not represented as active by this notice. Each remains unavailable for a Customer unless expressly ordered and independently cleared by the applicable provider, security, privacy, transfer and operational gates.
2. AI-generated content and human approval
2.1 Gated email or message content is not sent without a human decision. The Approval Gate records a bounded, redacted action summary, reviewer identity, decision time and payload digest. Before approving, the operator must open and review the full draft in its source surface and verify the recipient and material external effect, as clause 9.6 of the Master Services Agreement requires. Sensitive full content is not promised in the approval event, and release evidence proving that the digest binds the decision to the executed action and that later mutation is rejected remains outstanding.
2.1A Approval gating applies to Outbound Actions as defined in clause 1.1 of the Master Services Agreement — the sending of any email or email draft from a Connected Account, the sending of a message on an Enabled messaging channel to any person other than the Customer's operator user, and any further category expressly listed as approval-gated in Section 10 of the Order Form. Other operations within Connected Accounts — for example calendar creation and updates (which may notify attendees), file organisation and permission changes, and the reading and indexing of connected sources — are performed on the Customer's instruction, configuration and granted scopes without a per-action human decision, and are not Outbound Actions. A spoken turn of an enabled Voice Agent within an inbound call is not an Outbound Action and cannot be pre-approved; every email or message the Voice Agent proposes as a result of a call is gated.
2.2 Human approval is a control, not a guarantee. It does not establish that every AI output is accurate, create a statutory exception or transfer the Vendor's legal obligations. The Customer remains responsible for reviewing content it approves and for any recipient or public disclosure required by its use case.
2.3 The current release has not yet established a content-bound machine-readable provenance marker or consistent visible AI label on every possible surface. The Vendor must not claim that all AI content is marked until a release-specific gap analysis, implementation and robustness tests are approved.
3. Voice agent and mandatory AI disclosure
3.1 The optional Voice Add-On is disabled by default and is not part of the core Service. Voice v1 is inbound-only. Outbound dialling, marketing calls, call transfer and emergency-call routing are disabled.
3.2 Before any substantive call content, an enabled Voice Agent must make a non-interruptible mandatory AI disclosure identifying itself as an AI system, identifying the Customer for whom it acts, explaining the applicable audio/transcription processing and providing the approved privacy route. If the caller does not affirmatively continue, or asks not to be transcribed, the Voice Agent is designed to end the call without collecting further substantive information and to give the approved non-voice contact route. This behaviour is subject to the release evidence gate in 3.3.
The following wording is reproduced from clause 5.5 of the Voice Add-On Addendum, which is the contractual source of the mandatory opening; where this notice and that clause differ, that clause governs.
Required opening script (English): "Hello, this is an AI assistant for [Customer name], not a human. This call is transcribed, and audio and conversation content are shared with our approved voice and language-model providers to operate the call. More information and a non-voice contact option are available at [approved privacy URL]. Do you wish to continue?"
The approved privacy URL announced in the script is supplied and maintained by the Customer as Controller and contains the Customer's own privacy information for call participants (Voice Add-On Addendum clause 5.5(d)); the Vendor announces it as supplied.
3.3 The mandatory elements may not be removed, obscured or reworded without a new legal and readiness review. The Voice Add-On may be activated only after the exact provider and telephony entities, account tier, processing locations, contracts, transfers, training and retention settings, deletion behaviour, disclosure implementation and release tests are recorded in signed evidence. The mandatory elements must be delivered in the principal language of the call and of the caller population approved in the Voice Readiness Record. A translated opening is a rewording and requires the same legal and readiness review before use. The script carries a version identifier recorded in the per-call disclosure log.
3.4 The Voice Agent cannot contact or route to emergency services (999/112) and must not be relied on in an emergency. If a caller reports an immediate threat to life, safety or property, the Voice Agent is designed to stop the ordinary flow and direct the caller to end the call and contact the emergency services themselves; it does not accept an emergency report or undertake to relay one. This behaviour is not configurable.
4. Limitations of AI outputs
4.1 AI outputs are probabilistic, not authoritative. They can be inaccurate, incomplete, out of date, fabricated, biased or inconsistent, and can misread context, tone or intent. A plausible output is not proof that its facts, calculations, citations or requested actions are correct.
4.2 AI output is not legal, tax, accounting, financial, medical or other professional advice. It must not be relied on as such without review by an appropriately qualified human professional.
4.3 The Customer must verify every material output before relying on it and every outbound action before approving it; operations that are not approval-gated are governed by the configuration and scopes the Customer sets. The allocation of contractual responsibility, warranties and liability appears only in the Customer's frozen executed agreement.
4.4 The Vendor does not intentionally use Customer content to train its own models. A statement about an upstream provider's training, retention, region, transfer mechanism or content marking is made only after account-specific configuration readback and contract evidence are filed in the relevant release and DPA record.
5. Language, updates and contact
5.1 English is the sole authoritative version of this notice. No Polish translation forms part of the public legal set. A translation may be published only after it is produced from the final approved English notice, professionally synchronised, legally reviewed, version/hash bound and released atomically with the English source. No translation can change the governing law or contract route. This section governs the language of this published notice and does not govern the language of the in-call disclosure required by section 3.
5.2 Questions may be sent through jakub@novus-point.com or the additional tested contact stated in the Customer's executed Order Form. No response-time claim applies unless activated under the SLA. Complaints about a Customer communication may also be raised with the Customer that controls the relevant Workspace.
Related information: Website Terms · Acceptable Use Policy · Privacy Notice · Service Level Agreement. This notice is published information about how LEM uses AI and is not part of your executed contract set; under MSA clause 2.5 the contract set prevails where they disagree.