AI Transparency Statement

Effective Date: September 9, 2026
Last Updated: September 9, 2026
Version: 1

This statement explains, in plain language, how artificial intelligence is used in the BotB2B platform operated by Bot B2B, Inc. ("BotB2B", "we", "us"), what it can and cannot do, and what we ask of the businesses that use it. It is incorporated into the Terms of Service and referenced by the Data Processing Addendum. It is written for three audiences: our customers, the people who talk to our customers' bots, and the employees of our customers whose work reports are processed.

1. What BotB2B is

BotB2B is a business platform. Our customers (companies and sole proprietors) use it to run AI assistants that answer their customers on messengers, websites and marketplaces; to keep a CRM, tasks, notes and knowledge bases; to run autonomous "AI Managers"; and to collect and review work reports from their teams. BotB2B does not build its own foundation models. We send requests to AI models operated by third-party Model Providers (listed in our Sub-processor List) and return their answers.

2. Where AI is used

3. How we tell people they are dealing with AI

• Bots do not pretend to be people. Our base instructions never tell a Bot to hide that it is automated, and Bots answer truthfully when asked whether they are human. Customers write their Bots' greetings and instructions; they must never instruct a Bot to claim to be a person or to deny being automated, and where the law of their End Users' location requires an explicit disclosure at the start of a conversation, they must include it (we provide ready texts). Responsibility for a Bot's instructions rests with the customer (Acceptable Use Policy, Section 4).
• Handover to a person. When a Bot cannot answer, or when an End User asks for a person, the Bot can escalate to a human operator, and the conversation shows which messages were written by a person.
• AI-generated media. Images, audio and video generated through the platform are not to be presented as authentic recordings of real events or people. Where the law requires labelling of AI-generated content, customers are responsible for the label.

4. What the AI does with personal data

• Purpose limitation. Personal data in conversations, CRM records and reports is used to provide the feature the customer turned on, and for nothing else. We do not use it to train shared models, to profile individuals across customers, or for advertising.
• Model Providers. Model Providers process requests under commercial API terms that prohibit training on customer data and limit retention to a short abuse-monitoring window (see the Sub-processor List). The international platform uses providers located in the United States, Canada and the European Union.
• Optional safeguards. On request we can enable for a workspace additional safeguards such as masking of contact data written in standard formats before it is sent to a Model Provider. These are best-effort measures with limits that we describe when they are enabled; they are not anonymization.
• Human review. Our staff do not read customers' conversations except in read-only mode for troubleshooting, on the customer's request, or when investigating abuse, as described in the Terms of Service.

5. Workforce Insights in detail

Because this feature processes information about employees, we describe it precisely.

What it uses. Work reports that employees submit voluntarily by text or voice through the team app, at times set by their employer; the tasks, goals and plans recorded in the Workspace; and, where the employer chooses to enter them, the employee's role, schedule, absences and cost of employment. Voice reports are converted to text by a speech-to-text model; the audio is kept only as long as needed for transcription and the text of the report is stored in the Workspace.

What it produces.

• A structured summary of each report: what was done, what was planned but not done, blockers, wins and learnings, and a self-assessed completion percentage.
• Optional self-rated energy and stress levels (1–10) that the employee may state in the report. These are the employee's own statements; the platform does not measure or infer them.
• Trends over time (for example completion percentage over the last weeks), a comparison of each employee's results with that employee's own typical level, and a signal when reported results stay below the employee's personal norm for several consecutive days. Results are described in terms of delivery (below, at or above the personal norm) and of the employee's own reported energy and stress figures; the system does not label an employee's psychological state.
• Summaries of complaints and suggestions mentioned in reports, grouped by topic (personal excerpts are hidden by default), and, only where the employer enables the optional "Collaboration & Alignment Map" (off by default outside Russia), mentions of colleagues classified as positive, negative or neutral based on the words used in the report, aggregated into a collaboration score between team members. Managers can see these summaries for their teams.
• Suggested questions and discussion points for the manager, and, for roles the employer marks as revenue-related, an allocation of the employee's cost across the tasks reported.

What it does not do.

• It does not analyze tone of voice, facial expressions, typing behavior, screen contents, location, browsing, private messages or biometric data, and it does not create voiceprints.
• It does not infer emotions, mood, health or psychological state from voice, face or physiology. Any energy or stress figure is self-reported.
• It does not make, recommend or automate decisions on hiring, pay, promotion, discipline or termination. Its outputs are information for a human manager, who decides.
• It does not compare employees with a global benchmark; the comparison baseline is the employee's own history and, where the employer enables it, the team average.
• It does not monitor people who are not Members of the Workspace.

What we ask of employers. Inform your employees before you use it, in the form the law of your jurisdiction requires (we provide a template notice); keep a human in every decision; do not use it to discriminate; and complete any impact assessment, policy or consultation your jurisdiction requires. Employees can see their own results and can tell their manager when they disagree with a summary; managers can correct or re-run a summary.

6. Limits of AI Output

AI models predict text and other content. They can be wrong, invent facts, misunderstand context, reflect biases in their training data, and produce different answers to the same question. Outputs are not verified by BotB2B and are not professional advice. Customers must review Output before relying on it, and must supervise Bots and AI Managers as they would a new employee. We continuously improve prompts, safeguards and the model catalogue, but no AI system is error-free.

7. Human oversight and control

Customers control which AI Models are used, what Bots may say, what tools AI Managers may use, which employees are included in Workforce Insights and who can see the results. Every AI request is recorded in the Workspace's Token ledger with the model used, so customers can audit usage. Customers can pause a Bot, stop an AI Manager, delete conversations, Knowledge Bases and reports, and delete the whole Workspace.

8. Regulatory notes

• United States. We support customers' compliance with state laws that require disclosure of automated agents (for example California, Maine, Utah and New Jersey), with employee-notice and anti-discrimination rules for AI in employment (for example Illinois, California and Colorado), and with the Federal Trade Commission's guidance against deceptive AI claims.
• Canada and Latin America. We support customers' obligations under PIPEDA and Quebec's Law 25, Brazil's LGPD, Mexico's LFPDPPP and other data protection laws, including transparency about automated processing.
• European Union (for future availability). The design described above is intended to keep Bots within the transparency obligations of Article 50 of the EU AI Act and to keep Workforce Insights free of the practices prohibited by Article 5 (in particular, no emotion recognition from biometric data). Whether Workforce Insights is a high-risk system under Annex III depends on how a deployer uses it; we will publish the applicable conformity information before offering the platform in the EU.

9. Questions and reports

If you believe a Bot or an AI Manager has behaved improperly, or if you are an employee with questions about how your reports are processed, contact your employer (the customer) first, as it controls the data. You may also write to us at [email protected]; we will answer within ten (10) business days and will escalate genuine safety issues immediately.

This statement will be updated whenever we add AI capabilities that materially change what is described here.