KlarBond AI Processing Notice

Effective date: 29 June 2026

KlarBond uses AI and machine-learning systems to help organisations review, compare, organise and draft business documents and internal knowledge.

This notice explains what AI-assisted processing does, what information may be processed, the limits of the technology and the responsibilities of customers and users.

It should be read with the KlarBond Privacy Policy, Terms of Service, Restricted Data Policy and Subprocessor List.

1. AI-assisted features

Depending on the selected feature, KlarBond may use AI or machine-learning systems to:

• extract and structure document text;
• identify headings, sections or document structure;
• classify sentences, paragraphs, sections or documents;
• create embeddings that represent the meaning of selected text;
• search for semantically related content;
• compare information within or across documents;
• identify possible contradictions, tensions, gaps, duplication or outdated material;
• extract evidence and link findings to source material;
• answer questions based on selected documents;
• support research and summarisation;
• generate reports, drafts or other structured outputs; and
• reanalyse content when a document is replaced.

OCR is not currently a live KlarBond feature.

2. Typical AI data flow

A typical workflow may involve the following steps:

1. A user uploads a document or authorises KlarBond to retrieve it from Google Drive, Microsoft OneDrive or Microsoft SharePoint.
2. KlarBond stores or accesses the authorised source and extracts text and structure.
3. The Service divides relevant content into sentences, blocks, sections or other units.
4. KlarBond selects the content needed for the requested feature.
5. KlarBond sends prompts, instructions, selected text and limited metadata to an approved commercial AI provider where required for the feature.
6. The provider returns embeddings, classifications, findings, summaries, answers or generated text.
7. KlarBond stores relevant results, source references, usage records and workflow history in the Customer's workspace.

A feature may need to process substantial portions of a document where that is necessary to compare, structure or analyse it.

3. Information that may be processed

AI-assisted processing may involve:

• uploaded or imported document text;
• document titles, filenames, source identifiers, section labels and other metadata;
• prompts, questions, review instructions and selected options;
• text units, blocks, document pairs or evidence excerpts;
• embeddings and similarity information;
• AI-generated findings, explanations, classifications, summaries, answers and drafts;
• source references connecting an Output to supporting material; and
• model, endpoint, request, usage, error and security metadata.

Customer Content may contain ordinary personal data or confidential business information if the Customer chooses to include it.

4. External AI providers

KlarBond uses commercial AI providers only. KlarBond does not use consumer AI accounts to process Customer Content.

Before KlarBond onboards a customer to an AI-assisted feature that sends Customer Content to an external AI provider, KlarBond will identify the relevant provider or provider category in the Subprocessor List, customer agreement or other applicable customer-facing notice.

KlarBond may add or replace an AI provider after appropriate legal, privacy, security and technical review and any required customer notice. A provider may have different products, models, endpoints, storage features and retention settings, so the exact position is described by provider rather than assumed across all AI services.

5. Model training and product improvement

KlarBond does not use Customer Content to train or fine-tune a general-purpose AI model.

KlarBond does not permit an external AI provider to use Customer Content for model training unless:

• the Customer has expressly opted in through a separate written agreement signed by an authorised representative;
• the purpose, information, retention and safeguards are clearly described; and
• the use is lawful and consistent with the Customer's own obligations.

Ordinary use of the Service, acceptance of these notices, provision of support information or general product feedback is not an opt-in to model training.

KlarBond may use non-content operational information, such as feature usage, request volumes, error rates and security events, to operate and improve the Service. KlarBond will not use document text, prompts or Outputs for model evaluation or product improvement unless permitted by the Customer agreement and applicable law.

6. Provider retention and logs

Commercial AI providers may retain limited request, response or metadata information for service delivery, security, abuse prevention, support, reliability or legal compliance according to the applicable commercial terms and configuration. For some providers and products, this may include short-term safety or abuse-monitoring logs. Other products may store chat history, files, application state, cached prompts or audit logs for longer where those features require storage or where the customer enables them.

KlarBond's standard approach is to:

• disclose the relevant provider and retention position in the Subprocessor List, customer agreement or other applicable customer-facing notice before the provider processes Customer Content for an onboarded customer;
• disable optional provider data sharing or model-improvement settings for Customer Content where those settings are available;
• avoid provider-side file, conversation or vector-store features where KlarBond does not need them for the selected feature;
• assess each materially different endpoint, model, cache, file store, vector store, batch feature, fine-tuning feature or external tool before use; and
• use reduced-retention or zero-data-retention options where available, appropriate and commercially agreed.

Zero-data-retention is not a default commitment unless a separate written agreement or published provider configuration says it applies to the relevant feature. Transparency and no-training controls apply whether standard, reduced or zero-retention provider settings are used.

7. Documents belonging to clients and other third parties

Customers may submit documents received from clients, employees, suppliers or other third parties only where they have the right and authority to do so.

Before submitting third-party documents, the Customer must ensure that:

• the intended use is authorised;
• any client, controller, employer or licensor has provided required instructions or permission;
• affected individuals have received required privacy information;
• there is a valid lawful basis for personal-data processing;
• confidentiality and contractual restrictions have been considered; and
• the information complies with the Restricted Data Policy.

Where a Customer acts as a processor for its own client, the Customer must ensure that it may appoint KlarBond and KlarBond's approved providers as subprocessors.

8. Restricted information

During the closed beta, users must not submit:

• health or other special-category personal data;
• criminal-offence data;
• children's or vulnerable-person data;
• safeguarding records;
• legally privileged material;
• identity documents or detailed financial account data;
• payment-card authentication data;
• passwords, private keys, access tokens or other secrets;
• classified or export-controlled information; or
• information that may not be processed by an external cloud or AI provider.

The full restrictions are stated in the Restricted Data Policy. KlarBond does not guarantee that it will detect restricted information before it is processed.

KlarBond may consider a future enterprise exception only through separate written assessment and agreement. Access to the closed beta does not create any right to an exception.

9. Prompt injection and document manipulation

Uploaded and connected documents are untrusted input. A document can contain hidden, misleading or malicious text intended to influence an AI system, reveal information or alter an Output. This is commonly described as prompt injection or source poisoning.

KlarBond uses controls intended to reduce this risk, including separating system instructions from document content, limiting model permissions, scoping retrieval, using structured outputs, preserving source references and requiring human review.

These controls cannot guarantee that every malicious instruction or manipulated Output will be prevented.

KlarBond does not allow an AI Output by itself to execute code, disclose credentials, change permissions, contact third parties, make payments or take another external action without separate approved controls.

10. Human review and accuracy

AI-assisted Outputs may be incomplete, inaccurate, outdated, biased, misleading, inconsistent or unsuitable for a particular purpose. Similar prompts may produce different Outputs.

Users must:

• inspect the source material;
• verify important facts, quotations and calculations;
• apply professional judgement;
• correct or reject unsuitable Outputs; and
• avoid presenting Outputs as verified advice or fact unless independently checked by a suitably qualified person.

KlarBond does not provide legal, compliance, financial, medical, tax, employment, safeguarding or other professional advice.

11. Decisions about individuals

KlarBond does not make solely automated decisions that produce legal effects or similarly significant effects on individuals.

Customers must not use KlarBond as the sole or determinative basis for employment, recruitment, credit, insurance, education, healthcare, legal services, public benefits, safeguarding, immigration or other decisions about a person's rights, eligibility, access or treatment.

12. Customer controls

Available controls may vary by plan, product stage, feature and technical feasibility.

KlarBond may offer different AI processing options over time, such as standard AI processing, restricted AI processing, customer-approved provider settings or no external AI processing for selected use cases.

Only controls expressly shown in the Service or agreed in writing are available. This notice does not promise a no-external-AI mode, customer-selected model, regional model routing, reduced-retention mode or zero-data-retention mode unless that capability is stated in the applicable agreement or product configuration.

13. Changes

KlarBond may update this notice when AI providers, endpoints, models, data flows, retention configurations, risks or safeguards change.

14. Contact

KlarBond Ltd
Company number: 17300261
ICO registration number: ZC182066
Registered office: 5th Floor, 167–169 Great Portland Street, London, W1W 5PF, United Kingdom
Privacy: privacy@klarbond.com
Legal: legal@klarbond.com