This FAQ answers the security and privacy questions customers ask most about Agent One Teams and Audiences: what data the AI uses, which models power it, where data is processed, how long it is kept, and how you stay in control.
In this article:
Personal data and PII
Is personal data included in the system inputs?
No. Agent One doesn't retrieve personal data, such as an email address, user ID, or IP address, from the Unified Customer Database (UCD) or other systems to send to generative AI (GenAI). They build inputs from anonymous or aggregated context only.
Is personally identifiable information (PII) ever sent to OpenAI or another third party?
By default, no. If a user manually enters personal information in a prompt, such as their name and email address, that input can be sent to large language model (LLM) providers.
Are user prompts logged, and who can see them?
Insider One can log prompt inputs and outputs for system diagnostics. Only authorized internal teams can access the logs.
Are behavioral data, IDs, phone numbers, or emails used in prompts?
No. The system doesn't inject them. The only exception is data a user enters manually.
Is data transferred outside Turkey?
No personal data is transferred outside Turkey. In line with the Turkish Personal Data Protection Law (KVKK), the system sends no personal data to GenAI and uses only anonymized or non-personal data. Insider One processes non-personal data on AWS Ireland (eu-west-1). For extreme cases where personal data might be involved, client-side PII filters and masking are available.
Under GDPR and KVKK terminology, anonymized data is fully de-identified and falls outside the scope of data protection laws. Pseudonymized data still relates to an individual, so the law treats it as personal data.
How does the system prevent the identification of individuals?
The system doesn't share or use user data, so it can't identify individuals.
Model usage and AI transparency
Which AI models does Insider One use?
OpenAI GPT models, through OpenAI and Azure OpenAI/Anthropic Claude models, through Amazon Bedrock/Insider One's proprietary machine learning (ML) models for Decision Agent and personalization
Which AI-related software does Insider One use?
Insider One uses OpenAI, Azure OpenAI, and Amazon Bedrock.
Does Insider One train or fine-tune these models?
No. Insider One uses all LLMs as they are, through inference. It applies prompt engineering and context retrieval (retrieval-augmented generation, or RAG), but never fine-tunes a model with customer or user data.
Do the models learn from user data?
No. Insider One uses zero-retention APIs, so OpenAI and AWS don't retain or learn from the prompts Insider One sends.
How does Insider One ensure the accuracy of outputs?
Insider One manages output quality with:
System-level input and output validation.
Toxicity filtering through OpenAI and AWS APIs.
Manual prompt quality assurance, jailbreak tests, and end-to-end automated testing.
What are the AI models used for?
GPT models generate text and images, and act as the reasoning engine for Smart Segment and Smart Journey.
Claude Haiku powers the Support Agent and the Shopping Agent, because Insider One's infrastructure runs on AWS.
Data retention
How long are prompts stored?
Insider One doesn't store prompt data permanently. OpenAI and AWS use zero-retention endpoints. Insider One keeps only anonymized logs for diagnostics.
How does Insider One handle log retention and access for prompts?
Insider One sends all prompts through zero-retention endpoints (OpenAI and AWS Bedrock), so no data is stored or used for model training. It keeps anonymized prompt logs for diagnostics only, and these logs never include PII.
Role-based access control (RBAC) strictly limits access to the logs, and every access is audit-logged. The Insider One AI Platform team manages configuration and access.
How does Insider One delete and track product usage data?
Insider One deletes usage data as part of its data disposal process. It maintains deletion logs in line with its Data Processing Agreement (DPA) and retention policy.
System security and access control
Who can access the AI system?
Only authorized services and team members. Insider One manages access through RBAC and identity and access management (IAM) roles.
Who can view prompt inputs and outputs?
Only internal services and teams with specific permissions can access the logs, and the logs are anonymized.
Does Insider One test for security vulnerabilities?
Yes. Insider One runs:
Prompt jailbreaking tests.
Manual edge-case prompt reviews.
End-to-end automated tests.
Penetration testing and security reviews as part of the SSDLC.
How does Insider One apply IAM to model usage?
All model interactions run through secured service accounts with scoped API keys, authenticated requests, and allowlisted access controls.
Is the AI software proprietary or open source?
Insider One uses commercial (proprietary) AI models and services, not open-source LLMs.
Does Insider One log and monitor the AI system's activity?
Yes. Insider One logs all interactions.
Decision-making, traceability, and audits
Does Insider One check for harmful or misleading content?
Yes. OpenAI and AWS content filters and Insider One's own validation controllers filter AI responses.
Are AI outputs manually reviewed?
Yes. You review and approve AI-generated content before it goes live. The AI never acts without user validation.
Can I choose my own model?
Insider One selects the model based on the use case. For example, GPT models handle content, Claude handles analytics, and Insider One's own ML handles predictions. External model selection isn't currently supported, but you can customize prompts and outputs if needed.
Was the AI pre-trained, and on what data?
Insider One uses ready-made LLMs.
Does Insider One control for bias and discrimination?
Yes. Insider One AI products don't use personal identifiers, such as race, gender, or political orientation, during training or inference. ML-based predictions use only behavioral event data, such as clicks, purchases, and page views.
Third parties and subprocessors
Which third-party services does Insider One use?
OpenAI and Azure OpenAI for LLM inference/Amazon Bedrock for Claude models and infrastructure.
Is data shared with these providers?
Insider One shares only anonymous or aggregate data.
Does Insider One have subprocessor agreements and DPAs?
Yes. Insider One has formal DPAs with all subprocessors, including OpenAI and Azure. All subprocessors follow data protection regulations.
Does Insider One use an enterprise instance for the AI?
Yes. Insider One hosts its AI services on enterprise-level infrastructure within AWS, and every model it uses is enterprise-level.
Functionality and limitations
Do the AI systems act autonomously?
The assistant agents don't: every action that changes a campaign, segment, journey, or setting needs your confirmation. The Decision Agent acts within a boundary you set: it only picks among variants you approved, inside journeys you launched, on channels you enabled. It never sends content you haven't approved.
Which infrastructure runs the AI?
The GenAI modules use OpenAI GPT models. The predictive modules use Insider One's own ML, hosted on AWS. For details, see How Does Insider Manage ML Tasks and Machine Learning Supported Predictive Segments.