Responsible AI Policy

1. Why this policy exists

Most AI policies are written by companies that have never run a shift.

In2ition AI was built by operators. Twenty-plus years on the frontline. Hundreds of retail locations. Thousands of employees. Millions of customer conversations. We know what it feels like when a system grades your work and nobody can explain why. We know what it costs when a customer gets a wrong answer at the counter. We built this platform because that work deserved better tools - not because that work deserved to disappear.

That's the premise, and it's the whole thing: In2ition builds human-first AI.

Always-On Intelligence™ exists to make frontline people better at their jobs. To catch the calls nobody could get to. To coach in the moment instead of once a quarter. To give a rep in week two the pattern recognition of a rep in year five. Every design decision in this platform runs through that filter.

This policy is how we hold ourselves to it. It is not a statement of aspiration. Every commitment below is something a customer can verify, a security reviewer can test, or a contract can enforce.

2. Definitions

AI Employee - a configured agent deployed on the In2ition platform to perform a defined role (inbound support, outbound sales, appointment reminders, training delivery, candidate screening).

Customer Data - data supplied by, generated for, or captured on behalf of a customer, including call recordings, transcripts, summaries, scores, and coaching artifacts.

Configuration - the knowledge, scripts, escalation rules, and guardrails a customer sets for an AI Employee. Configuration is customer-controlled.

Consequential decision - a decision that materially affects a person's employment, compensation, eligibility, legal standing, or access to a service.

3. Our commitments

3.1 Human first

AI on this platform assists people, extends people, and escalates to people. It does not replace human judgment on decisions that matter to a person's livelihood or legal standing.

In practice:

• Every AI Employee has a defined escalation path to a human. Escalation is not optional and cannot be configured away.
• When an agent encounters a case outside its configured knowledge, it escalates rather than improvises.
• Deflection targets are a measure of what humans are freed from, not a measure of how many humans are removed. We say this to every customer during scoping, and we say it here.

3.2 People know they are talking to AI

No one should have to figure out whether they're speaking to a machine.

In practice:

• AI Employees identify as AI at the start of an interaction, per customer-configured disclosure language.
• An agent will never deny being an AI when asked. This is enforced at the platform level, not left to configuration.
• Recording disclosure and consent language is configurable to the strictest applicable jurisdiction, including all-party consent states.
• Our in-app recorder requires affirmative action by the user plus an audible disclosure layer before capture begins.

3.3 Grounded, not improvised

An agent that makes things up is worse than no agent.

In practice:

• Agents answer from configured knowledge sources, not from open-ended model recall.
• Compliance-critical language can be locked verbatim so it is delivered exactly as written, every time.
• Uncertain cases escalate to a human instead of guessing.
• Every interaction is scored by our QA layer, with the reasoning and the source moment attached - not a black-box grade.
• Customers control the knowledge. If a source is wrong, the answer will be wrong. We build the tooling to find that fast; we do not claim to be the arbiter of a customer's own program facts.

3.4 Your data is yours

In practice:

• We do not train on Customer Data. Not our models, not third-party models, no carve-outs. This is contractual, not a setting. It is written into our master agreement.
• Customers own their data. In2ition holds only the limited license required to operate the platform.
• We do not sell or license Customer Data. Ever.
• Where we produce aggregated and anonymized statistics (never identifiable, never customer-attributed), the permitted uses are limited and stated in the agreement, not buried in a policy page.
• Deletion and data-subject requests are supported with a defined fulfillment window. Submission procedures are described in our Privacy Policy.

3.5 Privacy is a default, not a feature

In practice:

• PII redaction is available across transcripts and stored artifacts.
• We collect what the interaction requires and no more. We do not request or store payment card data.
• We maintain a current sub-processor list, available to any customer on request.
• Access to Customer Data inside In2ition is role-based, logged, and limited to personnel who need it to deliver service.

3.6 AI does not make employment decisions

This is where "human-first" either means something or it doesn't.

Interaction Coaching, AI Training, and Employee Engagement produce scores, transcripts, coaching rationales, sentiment signals, and completion data. These are inputs to a human manager. They are not verdicts.

In practice:

• The platform does not take, recommend, or trigger adverse employment action. There is no configuration that enables it.
• Every score is explainable and traceable to timestamped moments in the underlying recording and transcript. A manager can see exactly what produced a number, and so can the employee.
• We support customer-run bias and outcome testing and will provide the documentation a customer needs for it.
• Where a customer operates under a law governing automated decision-making in employment, the employer is the decision-maker under those laws. Our obligation is to make their compliance possible: notice-supporting documentation, human-review pathways, correction workflows, and explainable outputs. We provide those.
• Coaching should develop people. If a deployment is being scoped to build a termination file, we will say so out loud and decline to configure it that way.

3.7 Accountable and auditable

In practice:

• Agentic actions are logged. What the agent did, when, on what input, under which configuration version.
• Model changes, prompt changes, and configuration changes are version-controlled.
• We notify customers of material changes to model providers or platform behavior that could affect their deployment.
• We investigate reported inaccuracies, document root cause, and report back to the customer.

4. What we will not build

Naming the line is the point. These are refusals, not roadmap gaps.

• Covert AI. No agent that conceals its nature when asked.
• Automated adverse action. No feature that fires, disciplines, demotes, or docks a person without a human decision-maker.
• Voiceprint identification or biometric surveillance. We transcribe and analyze conversations. We do not build systems to identify individuals by biometric signature.
• Impersonation of a real, named person without that person's documented consent.
• Manipulative agents. No manufactured urgency, no exploitation of confusion or vulnerability, no dark patterns run at conversational speed.
• Unsupervised deployment into legal or safety jeopardy. Where a wrong answer creates legal exposure or physical risk to a person, a human stays in the loop. We will not configure around that.
• Selling or licensing Customer Data, in any form, to anyone.

5. How this is governed

Accountability. The CEO and CTO own this policy jointly. It is not delegated to a committee that meets twice a year.

Review gate. New agent types and material capability changes go through an AI review covering intended use, failure modes, escalation design, disclosure requirements, and data handling - before deployment, not after.

Framework alignment. We operate the govern / map / measure / manage loop of the NIST AI Risk Management Framework. In2ition runs on cloud infrastructure providers that maintain SOC 2 Type II attestations, with encryption in transit and at rest.

Model and sub-processor governance. Foundation models and infrastructure providers are selected for capability, security posture, and data-handling terms consistent with this policy. We do not silently swap the model behind a production agent.

Incidents. Material accuracy failures, security events, and privacy incidents are investigated, documented, and reported to affected customers under the timelines in their agreement.

6. Shared responsibility

Trust breaks when nobody knows who owns what. Here's the split.

Customers agree to use the platform consistent with this policy and with applicable law. Deployments that conflict with Section 4 are outside the scope of what we will support.

7. Raising a concern

Any customer, employee, candidate, or member of the public who interacted with an AI Employee can raise a concern about AI behavior on this platform.

info@in2ition.ai

Concerns are reviewed by a human. We respond. We do not route ethics reports to a bot, which should go without saying and apparently doesn't.

8. Changes to this policy

This policy is reviewed at least annually and whenever platform capability changes materially. Versions are dated and retained. Material changes are communicated to active customers.

In2ition AI is the Always-On Intelligence™ platform for frontline businesses - conversational intelligence for your revenue layer. Human-first by design. See our Trust Center for security and infrastructure detail.