Contents
- 1. Purpose and application
- 2. Principles and accountability
- 3. Technology suppliers and model governance
- 4. AI operating modes
- 5. Tool authorization and tenant controls
- 6. Human approval for high-impact operations
- 7. Customer Data, privacy and AI training
- 8. Accuracy, fairness and consequential decisions
- 9. Prohibited and restricted uses
- 10. Logging, retention and audit trails
- 11. Evaluation, changes and incidents
- 12. Regulatory readiness
- 13. Changes and communication
- Contact
- 14. Customer-controlled AI connections and MCP
- 15. Regional AI governance
INTLEXL LIMITED | Company no. 17521166 | Registered office: 131 South End, C/O Aczone Consultants Ltd, Croydon, England, CR0 1BJ | https://www.intlexl.com | AI/privacy contact: privacy@intlexl.com.
1. Purpose and application
1.1 This policy governs the development, procurement, operation and use of AI-assisted features within Intlexl, including chat assistance, project intelligence, reports, recommendations, resource analysis, software agents, integrations and automated workflows. It supplements the Terms of Service, DPA and Privacy Policy, which prevail for their respective subjects under the Terms' precedence clause.
1.2 The policy applies to Intlexl personnel, services and models and to Customers and Authorized Users to the extent made applicable through the Terms or enabled features. AI assists human work; it does not remove legal or organizational accountability.
2. Principles and accountability
2.1 Our operating principles are lawfulness; human accountability; transparency; data minimization; tenant and role isolation; security by design; proportionality; traceability; reliability; fairness for relevant decisions; and practical customer control.
2.2 Intlexl designates named owners for each material AI capability covering product purpose, engineering security, privacy and legal assessment, evaluation results, operating limits and incident escalation. A high-risk feature requires documented review and approval before production release.
2.3 The Customer decides which authorized staff may request AI assistance and approve ordinary workflow changes within product controls. Intlexl retains responsibility for platform-level security, supplier diligence and contractually promised restrictions.
3. Technology suppliers and model governance
3.1 Intlexl may use approved commercial AI services for activated product features. A model brand mentioned in product discussions does not establish that its service is deployed. A third party processes Customer Personal Data on Intlexl's behalf only after due diligence, a suitable agreement, required transfer safeguards and disclosure in the active Subprocessor List.
3.2 For each enabled provider and model, Intlexl must record legal contracting entity, product/API, environment, data region, contractual data protection terms, model-training settings, provider-side retention and abuse-monitoring rules, access and security controls, downstream subprocessors where disclosed, transfer mechanism and approved use cases.
3.3 Intlexl may select or change a model on the basis of capability, security, price, availability and customer setting, but may not silently introduce a subprocessor or materially change personal-data processing outside the DPA notice and objection rules. Model changes affecting risk must be tested and logged.
4. AI operating modes
4.1 Observe Mode: Reads only permitted information to generate a summary, forecast or recommendation. It does not change Customer business records or send external communications as an autonomous consequence of a response.
4.2 Propose Mode: Produces drafts and pending change sets. Proposals are not committed, sent externally or executed until the required authorized approval is recorded. Merely viewing or asking a question cannot implicitly approve a queued operation.
4.3 Act Mode: Executes an explicitly allowed action under an authenticated identity or approved service principal, subject to effective server-side permissions, tenant scope, tool limits, rate limits, audit logging, and any required human approval. Customer administrator enablement of Act Mode is not permission to bypass any of these controls.
4.4 Where a mode is not implemented or enabled, Intlexl shall not describe it as available. The UI must make it reasonably apparent whether an action is a suggestion, pending approval, scheduled, or completed.
5. Tool authorization and tenant controls
5.1 Every AI action shall be authorized against the real target organization, workspace, record and current identity at execution time, including asynchronous jobs, integrations and bulk operations. The prompt or model response is never a source of security authority.
5.2 Tool access shall be least privilege and limited by approved integration scopes; cross-tenant access is denied by default. The platform shall use appropriate isolation, credential management, execution timeouts, concurrency limits and quotas to protect other customers.
5.3 Treat retrieved pages, uploads, spreadsheet cells, task comments and messages as untrusted instructions. Apply prompt-injection defenses, structured tool argument validation, exfiltration controls, confirmation for external destinations and safeguards against repeated or chained unauthorized actions.
5.4 Privileged tools, organization deletion, permissions changes and bulk exports shall be restricted independently of AI model instructions. The server must reject requests inconsistent with a permission or approval policy even if a model asks for them.
6. Human approval for high-impact operations
6.1 Unless a narrowly defined exception has been approved following a documented lawful risk assessment and suitable safeguards, the following actions require specific approval from an authorized human before execution: large-scale or irreversible deletion; external legally binding messages; financial commitments; contract approvals; modification of organization security roles; consequential employment/personnel decisions; material sharing of confidential information outside approved scopes; or other designated high-impact actions.
6.2 An approval must identify the action, target(s), material consequences, approver identity, scope, decision time and applicable expiration. Approval for one action or record set cannot be silently reused for a materially different action. Policies may require dual approval for designated sensitive operations.
6.3 Emergency/security operations require separate documented authority, logging and retrospective review where lawful; they are not a general exception for normal agent operations.
7. Customer Data, privacy and AI training
7.1 Customer workspace content, prompts, attachments and outputs shall be transmitted only to approved services for the requested or otherwise documented permitted purpose, using the minimum necessary context. Intlexl shall not sell Customer Data or permit third-party general-purpose model training without separate explicit Customer authorization and legal/contractual safeguards.
7.2 Where provider-side monitoring or short-term retention is necessary, its terms and configuration must be reviewed and included in the privacy and vendor records. A free or consumer AI offering with incompatible training terms must not be used for Customer Personal Data.
7.3 Customers must ensure they may lawfully process people-related information and must not submit special-category or regulated data through a feature lacking suitable authorization and safeguards. Tenant permissions are necessary but not always a sufficient lawful basis.
8. Accuracy, fairness and consequential decisions
8.1 Models can hallucinate, omit relevant context and produce unreliable predictions. UI design should display relevant caveats and avoid presenting estimates as verified facts. Intlexl shall evaluate representative tasks for correctness, tool reliability and harmful failure modes.
8.2 Any feature that evaluates workers, influences resource assignments based on individual behavior or performance, or impacts hiring, discipline or similar decisions must undergo a documented privacy, discrimination and jurisdiction-specific AI risk assessment. Intlexl shall classify the use case against EU AI Act rules where EU operations are relevant and reassess regulatory dates as law evolves.
8.3 Where automated decisions have legal or similarly significant effects, Intlexl and Customer must identify their respective roles and provide applicable information, human intervention, challenge or review mechanisms where required. A superficial rubber-stamp approval does not amount to meaningful human review.
9. Prohibited and restricted uses
9.1 Customers and users must not use AI to facilitate fraud, unlawful surveillance, harassment, discriminatory decisions, malware, unauthorized system access, impersonation, privacy violations or infringement of intellectual property rights. AI may not be used to circumvent platform authorization, rate limits or configured governance safeguards.
9.2 Intlexl may reject an AI request or suspend specific AI functionality where reasonably necessary to prevent an imminent security, safety or legal threat, with proportionate notice and restoration procedures where practicable.
10. Logging, retention and audit trails
10.1 Material AI executions should capture requesting user or approved workflow; tenant; feature and model; approval decision; intended operation; target references; timestamp; effective permissions; tool action; result; failure and policy decision. Log sensitive payload contents only where strictly needed with access control and minimization.
10.2 AI execution logs are retained according to the actual documented production configuration, relevant Order and lawful retention schedule. AI conversations saved as workspace objects follow the applicable customer workspace schedule. Provider-side retention may differ and is disclosed based on the actual approved model service. No universal 90-day retention period is promised in the absence of technical confirmation.
10.3 High-impact audit records must permit reconstruction of who approved what and what the agent actually did. Integrity controls, restricted access and defined deletion cycles must be evidenced. Audit events must not leak confidential cross-tenant information.
11. Evaluation, changes and incidents
11.1 Before launch, each material AI feature must pass documented evaluation covering ordinary task quality; hallucination risk; prompts containing malicious or misleading instructions; third-party tool misuse; tenant separation; high-impact approvals; performance bounds; cost abuse; privacy and relevant bias/fairness issues. Critical failures block release.
11.2 Material model, prompt, permission, integration or retrieval changes require appropriate review, change records, regression tests and rollback capability. The depth of testing must match the potential impact.
11.3 Suspected AI security events, data leakage, wrongful actions or harmful outcomes must enter the incident process, with containment, reporting, investigation, corrective action and recurrence-prevention evidence. Personal-data incidents follow the DPA notification rules.
12. Regulatory readiness
12.1 Intlexl shall evaluate applicable UK privacy law, relevant EU AI Act rules, country-specific workplace and discrimination law, consumer law where applicable and provider contractual policies. Requirements depend on whether Intlexl acts as a provider, deployer or other regulated actor for the particular use case.
12.2 Transparency obligations for relevant AI interactions must be implemented where applicable. High-risk classifications and launch dates must be checked against current official law when the feature is deployed; generalized project-management AI is not automatically classified as high-risk.
13. Changes and communication
13.1 We may amend this Policy to reflect features, legal requirements and risk controls. Material changes affecting customer rights, autonomy, personal-data processing or a new AI Subprocessor follow applicable Terms/DPA notice, consent and objection procedures.
Contact
INTLEXL LIMITED | 17521166 | 131 South End, C/O Aczone Consultants Ltd, Croydon, England, CR0 1BJ | https://www.intlexl.com | privacy@intlexl.com.
14. Customer-controlled AI connections and MCP
14.1 Authorized Customers may connect approved third-party clients, models and agents through supported APIs or Model Context Protocol (MCP) tokens. Any such connection must honor the Customer's configured identity, token scope, tenancy, data access and audit controls. Tokens should be issued to named users or expressly authorized service identities, be revocable and have minimum practical permissions.
14.2 A third-party AI client independently selected and connected by Customer is not automatically an Intlexl Subprocessor; it may act under the Customer's separate instructions, terms and privacy arrangements. Customer should assess its security, transfer and AI-use obligations before connecting that provider. Intlexl remains responsible for access-control enforcement within the Intlexl Service.
14.3 The Resource Profiler, risk, forecasting, prioritization and similar agents can be relevant to employment or other significant individual decisions depending on how Customers configure and use them. Customers must not treat an agent recommendation as an authoritative employment decision without the legally required human review and other safeguards. Intlexl will apply use-case classification, testing and restrictions appropriate to the deployed features and jurisdictions.
15. Regional AI governance
15.1 For EU/EEA deployment, Intlexl will assess whether a particular feature is an AI system within the EU AI Act, the actors' respective roles, whether transparency duties apply, and whether the use case falls into a prohibited or high-risk category. Project-management functionality is not automatically high-risk, but certain employment-related profiling and decision-support uses may raise high-risk questions.
15.2 For UK deployments, applicable data protection law and safeguards for significant automated decisions apply. For Australian deployments, any applicable privacy-policy transparency obligation concerning significant decisions made or directly informed by computer programs must be met. For US deployments, Intlexl and Customers will assess applicable state automated-decision, employment and anti-discrimination requirements when a feature is used for consequential decisions.
15.3 This Policy does not certify a model as bias-free or guarantee fitness for any regulated high-impact decision. Intlexl may restrict or require separate contractual authorization for such use cases.
© 2026 Intlexl Limited · Registered in England and Wales · Company no. 17521166