TABLE OF CONTENTS

• 1. How does Freddy AI use my data?
• 2. How do I opt out of AI Training on my data?
• 3. Can I disable AI features if I choose not to use them?
• 4. How do I activate AI features after purchasing? 
• 5. How does Freddy AI protect your data? 
• 6. Is customer data ever used to train Freddy AI models?
• 7. Is customer data ever shared with third-party AI providers to train their models?
• 8. What features will customers lose when they opt out of training all AI models?
• 9. What features will customers lose when they opt out of training Collaborative Models?
• 10. How does Freddy AI prevent the generation of toxic, unsafe, or manipulated content?
• 11. How does Freddy AI ensure unbiased decision-making, answer, and action accuracy?
• 12. How does Freddy AI improve over time?
• 13. What models does Freddy AI use?
• 14. What is the difference between OpenAI and Azure OpenAI? 
• 15. By adopting Freddy AI features, will any of the data ever leave the region where the customer is hosted? 
• 16. Does using Freddy AI change my analysis of HIPAA compliance?

Understand how Freddy AI maintains data integrity and follows responsible AI practices through these comprehensive FAQs. These resources provide clarity on the safeguards in place to help you use Freddy AI with confidence.

These FAQs apply exclusively to Freshworks Freddy AI-branded products. They do not cover Device42 or FireHydrant products and services.

1. How does Freddy AI use my data?

Freddy AI takes a security-minded approach to developing and deploying AI features. Many of these features use a combination of model types to provide the best balance of quality and performance. Freddy AI uses three main types of AI models:

• Custom models: These models are hosted by Freshworks and may be trained on your organization’s data to improve your individual product experience.

• Collaborative models: These Freshworks-hosted models are trained on data from multiple users to improve model accuracy and performance for everyone.

• Third-party generative models: Freddy AI uses Microsoft Azure OpenAI for certain features. Under Azure’s terms, your data is not stored long-term by Microsoft, is not used for model training or fine-tuning, and is processed only to generate results.

How your data is used depends on which products and features you use.

You may opt out of AI features, including the use of your data for AI training, at any time by contacting support@freshworks.com.

2. How do I opt out of AI Training on my data?

Freshworks offers you freedom of choice in how your data is used by AI and in AI training of Freshworks models. It only uses your data for training if you use AI features. If you do not purchase or have any AI features enabled, your data is not used for AI training purposes. If you have enabled AI features, you may opt out of your data being used to train AI models in the following ways:

• Opting out of collaborative models: If you want to opt out of your data being used to train models trained on data from multiple users, you may do so by contacting support@freshworks.com. Once you opt out, you will continue to have access to Freddy AI features, though performance may be reduced since your data will not contribute to those shared model improvements.

• Opting out of all AI training: If you wish to opt out of your data being used to train both collaborative and custom models, you may do so by contacting support@freshworks.com. By doing so, you will disable AI features that rely on model training, such as those powered by collaborative and custom models. You will still have access to AI features that do not use your data for training, including those powered by third-party generative models like Microsoft Azure OpenAI.

Opting out of AI training has no effect on non-AI related Freshworks services.

3. Can I disable AI features if I choose not to use them?

You can turn individual AI features on or off at any time directly from your administrative console. These settings can be adjusted at your discretion, and changes take effect immediately.

4. How do I activate AI features after purchasing? 

Many Freshworks AI-powered features are not activated by default at the time of purchase. To activate them, your administrator must visit the console settings and turn on each desired feature within the product experience.

In some instances, you may also need to assign seats for AI features to your team members so they can use the features. Assigning seats is required for the features to be available to agents.

5. How does Freddy AI protect your data? 

Freddy AI is built on a foundation of enterprise-grade security and clear data practices. The architecture uses strict trust controls to help you meet security and compliance requirements. For more information, visit the Freshworks Trust Center.

AI guardrails guide Freddy AI to act in predictable and responsible ways. Data privacy, model explainability, and governance are built into the Freshworks AI trust framework to reduce risk and support responsible use across Freshworks services.

6. Is customer data ever used to train Freddy AI models?

Freddy AI uses data from consenting users to train many of the AI models that power, personalize, and improve AI product performance. Freddy AI uses data from consenting users for training two distinct types of models:

• Custom models: These are trained only on your data and used exclusively within your account.

• Collaborative models: These are trained on de-identified data from multiple users to enhance shared performance.

Notably, Freshworks never uses your data to train third-party hosted generative models.

Freshworks is constantly working to improve AI processes and services. Currently, some products utilize AI models trained on data that was not de-identified prior to training. This is limited to collaborative models for Freshsales and Freshmarketer. Freshworks is actively updating and retraining these models using only de-identified data to adhere to improved data security practices.

You can opt out of either type of model training at any time by contacting support@freshworks.com. Opting out may reduce the performance of some AI features. Once your opt-out is confirmed, your data will be removed from AI training sets without undue delay, in accordance with the Freshworks retention policy. For more details, refer to the Opt-Out Terms in the Freshworks supplemental terms and documentation.

7. Is customer data ever shared with third-party AI providers to train their models?

Freshworks does not share your data with any third-party providers for AI model training.

Freshworks believes your data should be used only to improve your experience with Freshworks products, and not sold, transferred, or used to benefit external AI systems.

8. What features will customers lose when they opt out of training all AI models?

When you opt out of all AI training, any features powered by custom or collaborative models are disabled. You will still have access to AI capabilities that do not rely on data training to operate.

After opting out, available functionality is limited to tasks such as summarization, translation, and certain routing capabilities. You will lose access to personalized insights, automations based on your data (such as knowledge base or ticket history), and any AI-driven recommendations or suggestions derived from your usage patterns.

9. What features will customers lose when they opt out of training Collaborative Models?

When you opt out of training collaborative models, your services remain unchanged. You may see slightly different performance in AI features powered by collaborative models after you opt out of AI training.

10. How does Freddy AI prevent the generation of toxic, unsafe, or manipulated content?

Freddy AI incorporates multiple layers of safeguards to ensure responses remain safe, responsible, and appropriate for enterprise use. These protections combine Microsoft Azure and AWS industry-standard safety systems with Freshworks’ own controls.

Content safety tools are applied in real time to both inputs and outputs. These filters use machine learning classifiers to detect and block hate speech, explicit material, violence, and self-harm content. They are configured at low severity thresholds to identify even subtle risks. These tools also help prevent out-of-context prompts from producing unintended results and reduce the likelihood of hallucinations.

To protect against manipulation attempts such as prompt injection or jailbreaking, Freddy AI uses prompt shields that analyze your queries for malicious intent before they reach the model. Together, these layers of defense keep Freddy AI aligned with its intended purpose of providing secure, contextually relevant AI experiences.

11. How does Freddy AI ensure unbiased decision-making, answer, and action accuracy?

Freddy AI is designed to minimize bias and deliver accurate, reliable responses through a multi-layered approach. Safeguards such as grounding, human review, and feedback reporting are built into every stage of Freshworks AI development to help identify and reduce potential bias.

Models are trained using domain-specific data and evaluated against real support scenarios, rather than relying on generic public data. The Freddy AI architecture also uses Retrieval-Augmented Generation (RAG), ensuring results are grounded in your own trusted data sources.

Custom models are built to respond only within the context of your data, helping to prevent irrelevant content or biased outcomes.

As with all forms of AI, you should also review the output of Freddy AI to ensure the content and results produced in your instance are relevant to you.

12. How does Freddy AI improve over time?

Freddy AI is continuously evolving to deliver better, more precise performance. The agile model framework allows Freshworks to evaluate, develop, train, and deploy new AI models quickly and seamlessly, without disrupting your product experience or requiring reconfiguration.

Beyond system improvements made by Freshworks, you play a key role in enhancing your own AI outcomes. By providing richer context, high-quality data, regular feedback, and well-maintained knowledge resources, you enable Freddy AI to generate more accurate, tailored insights for your business.

If you regularly update your knowledge bases and supply meaningful feedback, you will typically experience the best results and most relevant responses from Freddy AI.

13. What models does Freddy AI use?

Freddy AI combines both commercially available and proprietary in-house models to deliver secure, enterprise-grade AI experiences.

For certain generative capabilities, Freddy AI uses Microsoft Azure OpenAI models. These models operate under enterprise-grade contractual terms that ensure all data processed by Microsoft is handled securely, without retention or training. For more details on how Microsoft trains its models, please refer to Microsoft’s official disclosures.

Proprietary Freshworks models are trained solely on data from consenting users within the secured Freshworks AWS infrastructure, protected by strict access controls and privacy safeguards. As Freshworks re-trains AI models, training activity is documented with records that track the data sources, training dates, and retention timelines to maintain full traceability and accountability in alignment with Freshworks' security and compliance commitments.

If you prefer not to have your data used for AI training, you can either opt out of model training or disable AI features at any time.

14. What is the difference between OpenAI and Azure OpenAI? 

OpenAI is the organization that develops advanced language models such as GPT, Codex, and DALL-E. These models can be accessed directly through OpenAI or through Microsoft’s Azure OpenAI Service, which provides the same models within Microsoft’s enterprise-grade cloud infrastructure.

Azure OpenAI is co-developed and maintained in partnership with OpenAI, ensuring API compatibility and a seamless experience across both providers.

Freshworks partners with Azure OpenAI to leverage these advanced models with the added benefits of Microsoft’s security, compliance, private networking, regional availability, and responsible AI safeguards for your data.

15. By adopting Freddy AI features, will any of the data ever leave the region where the customer is hosted? 

If you are hosted in the U.S. or the E.U., your data remains within its respective region. For example, data from an E.U. hosted account never leaves the Freshworks E.U. data center.

If you are hosted in other regions, your data may be transferred to the Azure OpenAI service, which is hosted in the U.S. or another region. This data is never stored in that region; it is only transferred to generate AI inferences. As Azure OpenAI becomes available in your respective region, Freshworks will localize the service to your designated geographic region. These updates will be reflected in support documentation or release notes.

16. Does using Freddy AI change my analysis of HIPAA compliance?

Freddy AI processes your data with minimal adjustment and, through contractual agreements with third-party AI providers, aims to meet necessary obligations. Currently, the only third-party AI models receiving data in Freddy AI are hosted by Microsoft on its Azure cloud platform. Freshworks' use of Azure-hosted features is governed by the Microsoft Online Services Data Protection Addendum, including the Microsoft HIPAA Business Associate Agreement. For further details, consult the Health & Life Sciences section of Microsoft’s compliance documentation.

Ultimately, you are solely responsible for determining if there are additional considerations regarding your use of AI and your specific HIPAA obligations. Freshworks recommends consulting with your legal advisor if you have further questions or concerns about your particular situation.