Terms of Service — AI Psychological Safety Monitoring

Effective Date: 2 February 2026

Last Updated: 2 February 2026

These Terms of Service ("Terms") govern your access to and use of Empathyc's AI Psychological Safety Monitoring service (the "Service"), provided by Keido Labs Ltd ("we," "us," or "our").

⚠️ CRITICAL DISCLAIMER — READ BEFORE USE

Empathyc is a monitoring and alerting tool, NOT a crisis intervention service.

What we ARE:

A "smoke detector" for potentially harmful AI conversations
An automated monitoring system using AI-based analysis
An alerting platform that flags conversations for your review

What we are NOT:

A crisis response service
A mental health provider or counselor
A substitute for human judgment or professional crisis intervention
A guarantee that harmful conversations will be detected or prevented

By using this Service, you acknowledge and accept full responsibility for:

Reviewing alerts and deciding whether action is needed
Implementing your own crisis response protocols
Ensuring appropriate human oversight of AI conversations
Any outcomes resulting from conversations on your platform

If you require guaranteed crisis detection or response, DO NOT use this Service. Hire qualified mental health professionals and implement appropriate duty of care protocols.

1. Agreement to Terms

By accessing or using the Service, you agree to be bound by these Terms. If you do not agree, do not use the Service.

Company Details:

Company Name: Keido Labs Ltd
Company Number: 16805440
Registered Address: 31 Russell Street, Liverpool, England, L3 5LJ
Email: hello@keidolabs.com
Website: https://empathyc.co

2. Description of Service

2.1 What the Service Does

Empathyc monitors AI-powered conversations (customer support, coaching, companion bots, or other conversational AI systems) and analyzes them for psychological safety risks using automated AI-based evaluation.

Service Components:

Real-time conversation monitoring via API integration
Automated psychological safety analysis using multiple metrics (see Section 2.3)
Alert generation when potential risks are detected
Dashboard for reviewing flagged conversations and analysis results
Timeline view of full conversation context for human review

Integration Methods:

Direct API integration (you send conversations to our endpoint)
OAuth integrations with supported platforms (Intercom, Zendesk, etc.)

2.2 What the Service Does NOT Do

IMPORTANT: Please read this carefully to understand what we do NOT provide:

We do NOT:

Intervene in conversations - We do not stop, block, or modify conversations in real-time
Respond to crises - We do not contact users, call emergency services, or take any action in response to detected crises
Store personally identifiable information (PII) - We do not collect user names, emails, contact details, demographics, or location data (see Section 7)
Provide mental health services - We are not therapists, counselors, or mental health professionals
Guarantee detection - We cannot guarantee that all harmful conversations will be detected (see Section 2.4)
Replace human judgment - We provide signals and evidence, not decisions

Why we're structured this way:

By design, we do not collect PII (personally identifiable information about the individuals having conversations). This means:

We physically cannot contact users in crisis
We have no ability to intervene or respond
We cannot take action beyond alerting you (the Service customer)

This is intentional. We are a monitoring tool, like a smoke detector. Smoke detectors alert you to potential fire. They don't call the fire brigade. That's your responsibility.

Similarly, Empathyc alerts you to potential psychological safety risks. You decide whether to act. You contact the user if appropriate. You implement your crisis response protocols.

2.3 Psychological Safety Metrics

We analyze conversations using the following metrics (collectively, "Safety Metrics"):

Empathy Score (0-10): Measures emotion recognition, validation, and appropriate tone matching
Reliability Score (0-10): Measures accurate expectations, follow-through, and limitation disclosure
Consistency Score (0-10): Measures factual coherence, context retention, and logical flow
Crisis Detection Score (0-10): Detects direct or indirect indicators of suicidal ideation, self-harm intent, or severe hopelessness
Advice Safety Score (0-10): Detects inappropriate medical, legal, or financial advice
Boundary Safety Score (0-10): Detects inappropriate intimacy, manipulation, or dependency patterns
Composite Safety Score (0-10): Weighted aggregate score with threshold-based risk classification

Each metric is evaluated using AI-based analysis (see Section 2.4) and backed by peer-reviewed psychological and computational research (2024-2026 published studies).

Alert Triggers:

Alerts are generated when:

Crisis Detection Score indicates direct or indirect crisis language
Any Safety Metric falls below configurable thresholds you set
Multiple metrics indicate concerning patterns

You control alert thresholds and delivery (email, Slack webhooks, or dashboard-only).

2.4 AI-Based Analysis — Limitations and Accuracy

READ THIS SECTION CAREFULLY:

Our Safety Metrics are generated using AI-powered automated analysis, specifically a "LLM-as-a-judge" system. This means we use large language models (LLMs) from third-party providers (OpenAI, Anthropic, Google Gemini, and others) to evaluate conversation content against our clinical rubrics.

What this means for accuracy:

AI-based analysis is probabilistic, not deterministic. This means:

Analysis is based on statistical patterns, not human clinical judgment
Accuracy is estimated at 85-90% at best for crisis detection in optimal conditions
False positives WILL occur (alerts for conversations that are not actually risky)
False negatives WILL occur (failure to alert on some risky conversations)
Edge cases, novel phrasing, cultural context, and domain-specific language may reduce accuracy
Different AI models or evaluation runs may produce slightly different scores

False Positive vs. False Negative Trade-off:

Our system is intentionally tuned for high sensitivity (bias toward false positives over false negatives). This means:

We would rather alert you to 10 conversations that are fine than miss 1 conversation that is risky
You will receive alerts that, upon human review, you determine are not actually concerning
This is by design for safety-critical domains

What we do NOT guarantee:

We do NOT guarantee 100% detection of all crises or harmful conversations
We do NOT guarantee zero false positives or false negatives
We do NOT guarantee that our analysis matches human clinical judgment in all cases
We do NOT guarantee that every crisis indicator will be caught

What we DO guarantee:

We use peer-reviewed psychological frameworks and best-available AI models
We continuously improve our analysis methodology based on research and production data
We provide transparent reasoning and evidence quotes for each analysis
We make commercially reasonable efforts to maintain high detection quality

Your responsibility:

YOU MUST IMPLEMENT HUMAN REVIEW. The Service is designed to assist human decision-making, not replace it.

You are responsible for:

Reviewing alerts and making final determinations about risk
Using your own judgment, domain expertise, and context to interpret analysis results
Implementing appropriate escalation and crisis response protocols
NOT relying solely on automated analysis for decisions that could impact individual safety

We provide the signal. You make the decision.

2.5 Continuous Improvement and Methodology Changes

We continuously improve our AI models, evaluation rubrics, scoring algorithms, alert thresholds, and analysis methodology (collectively, "Evaluation Methodology").

We reserve the right to:

Modify our Evaluation Methodology at any time without prior notice
Retrain AI models using additional data
Adjust scoring thresholds and risk classifications
Update the underlying LLM providers or computational approaches
Change how Safety Metrics are calculated or weighted

What this means for you:

Analysis results for the same conversation may change over time as we improve models
Historical scores may not be directly comparable to current scores after methodology updates
We are NOT required to notify you of every change to our Evaluation Methodology
Improvements are made to enhance overall service quality and accuracy

3. Third-Party AI Model Processing

IMPORTANT: By using the Service, you consent to the following:

3.1 LLM Provider Processing

We use third-party large language model (LLM) providers to analyze conversation content. Currently, these providers include:

OpenAI (GPT-4, GPT-4o, and other models)
Anthropic (Claude models)
Google (Gemini models)
Other AI model providers as we continue to improve the Service

What this means:

Conversation data you send to Empathyc will be transmitted to these AI model providers for analysis
These providers process data according to their own privacy policies and terms of service
We use standard API agreements with data processing protections where available
We may change, add, or replace AI model providers at any time without prior notice

Your consent: By using the Service, you consent to your conversation data being processed by our current and future AI model providers for the purpose of providing Safety Metrics and alerts.

Your responsibility: You represent and warrant that you have obtained all necessary consents and have the legal right to share conversation data with us and our AI model providers.

3.2 Data for Service Improvement

We may use aggregated and anonymized conversation data to:

Improve AI analysis models and detection accuracy
Develop new Safety Metrics and features
Conduct research on conversational AI safety
Publish anonymized case studies or research findings

We will NOT:

Sell your data to third parties
Use your data for advertising or marketing to others
Share identifiable conversation data with competitors
Use your data in ways that could identify your customers or your organization

4. Your Responsibilities — Duty of Care

THIS SECTION IS CRITICAL. PLEASE READ CAREFULLY.

4.1 You Are Responsible for Crisis Response

Empathyc does not assume, and explicitly disclaims, any duty of care toward individuals having conversations on your platform.

You retain full responsibility for:

Implementing appropriate crisis response protocols
Deciding whether and how to respond to alerts
Identifying users from conversation IDs (we do not store PII)
Contacting users in crisis if appropriate
Involving emergency services, mental health professionals, or other responders as needed
Complying with applicable laws regarding duty of care, mandatory reporting, or other obligations

The Service is a tool to assist your crisis detection efforts. It does not replace your duty of care obligations.

4.2 Human Review is Mandatory

You MUST implement human review of flagged conversations.

You agree that:

Automated analysis results will be reviewed by qualified personnel before taking action
You will NOT rely solely on Empathyc's automated analysis for decisions that could significantly impact individual safety
You will use appropriate judgment, context, and expertise when interpreting alerts
You will implement internal policies and training for staff reviewing alerts

Dashboard for Human Review:

We provide a dashboard with full conversation timelines, Safety Metric scores, LLM reasoning, and evidence quotes specifically to enable human review. You are responsible for using it.

4.3 Knowing Your Users

By design, we do NOT collect PII. This means:

We store opaque conversation IDs that you control (e.g., "conv_12345")
We do NOT know who the individuals are, how to contact them, or any demographic information
You are responsible for mapping conversation IDs to user identities in your own system

Crisis Response Workflow (Your Responsibility):

Empathyc sends alert with conversation ID and analysis
You receive alert via email, Slack, or dashboard
You review the conversation context and Safety Metrics
You look up the conversation ID in your own system to identify the user
You decide whether action is needed based on your context and policies
You contact the user, involve professionals, or take other appropriate action

Intentional friction: The manual lookup step forces human review before PII access. This is by design for safety-critical decision-making.

4.4 Compliance with Applicable Laws

You are responsible for:

Complying with data protection laws (GDPR, UK DPA 2018, etc.) regarding conversation data
Obtaining necessary consents from your users to monitor conversations
Complying with mental health or crisis response regulations in your jurisdiction
Meeting any duty of care obligations under applicable law
Complying with industry-specific regulations (healthcare, coaching, education, etc.)

We are NOT responsible for ensuring your compliance with applicable laws. Consult your own legal advisors.

5. Privacy and Data Protection

5.1 What Data We Collect

Conversation Data (what we DO collect):

Full text of AI and user messages in conversations you send to us
Opaque conversation IDs (identifiers you control, e.g., "conv_abc123")
Timestamps of messages and conversations
Safety Metrics, analysis results, and alert history

What we do NOT collect:

Personally identifiable information (PII) about conversation participants
Names, email addresses, phone numbers, or contact details
Demographic information (age, gender, location, etc.)
IP addresses or device identifiers of conversation participants
Any data that would allow us to identify or contact individuals

Why this matters:

GDPR-friendly by design (we process conversation content, not PII)
You retain data sovereignty and control over user identities
Zero-knowledge monitoring (we analyze conversations without knowing who's having them)
Legal protection: we physically cannot take action on crises because we don't know who the user is

5.2 Data Retention

Active accounts: Conversation data and analysis results are retained for the duration of your subscription plus 30 days
Cancelled accounts: Data is deleted within 30 days of cancellation (unless legally required to retain)
Data export: You may request a data export (CSV/JSON) at any time by contacting us
Deletion requests: You may request deletion of specific conversations at any time

5.3 Data Security

We implement industry-standard security measures:

Encryption in transit (TLS 1.2+)
Encryption at rest (AES-256)
Multi-tenant architecture with tenant isolation
AWS hosting (Ireland region for EU/UK data residency)
Access controls and audit logging
Regular security updates and monitoring

However: No system is 100% secure. You use the Service at your own risk.

5.4 GDPR and Data Processing

For EU/UK customers:

We act as a Data Processor for conversation content
You act as the Data Controller
We comply with GDPR requirements, including processing data only on your instructions, implementing appropriate security, assisting with data subject rights requests, and notifying you of data breaches

Data Processing Agreement (DPA):

For enterprise customers requiring a separate DPA, contact hello@keidolabs.com.

6. Limitation of Liability

READ THIS SECTION CAREFULLY. IT LIMITS OUR LIABILITY.

6.1 Nature of the Service

Empathyc is a monitoring and alerting tool, not a crisis prevention service.

YOU ACKNOWLEDGE AND AGREE:

The Service is provided as a best-effort detection system, not a foolproof crisis prevention system
AI-based analysis has inherent limitations, inaccuracies, and failure modes (see Section 2.4)
False negatives (missed crises) and false positives (unnecessary alerts) WILL occur
We do NOT guarantee detection of all harmful conversations or crisis situations
We do NOT guarantee prevention of any harm, injury, or negative outcome
The Service is ONE tool among many for managing conversational AI safety

You are responsible for implementing comprehensive safety protocols, not relying solely on Empathyc.

6.2 Disclaimer of Warranties

TO THE MAXIMUM EXTENT PERMITTED BY LAW:

THE SERVICE IS PROVIDED "AS IS" WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO:

Merchantability or fitness for a particular purpose
Accuracy, completeness, or reliability of Safety Metrics or analysis results
Detection of all harmful conversations, crises, or safety risks
Prevention of harm, injury, or negative outcomes
Uninterrupted or error-free operation
Security of data transmission or storage

6.3 Limitation of Liability — Cap on Damages

TO THE MAXIMUM EXTENT PERMITTED BY LAW:

Our total liability for any claims arising from your use of the Service (including but not limited to claims related to missed crisis detection, false positives, false negatives, service failures, data breaches, or any other issue) is limited to:

The amount you paid us in the 12 months prior to the claim (or £100 if you paid nothing).

6.4 Exclusion of Consequential Damages

WE ARE NOT LIABLE FOR:

Indirect, incidental, special, exemplary, punitive, or consequential damages
Lost profits, revenue, data, or business opportunities
Damage to reputation or goodwill
Costs of substitute services
Harm, injury, or death of any individual (whether or not related to missed crisis detection)
Claims by third parties (including your customers or their families) related to crisis response or lack thereof
Legal claims, lawsuits, or settlements resulting from conversations on your platform

6.5 Exclusions to Limitations

The above limitations do NOT apply to:

Death or personal injury caused by our negligence
Fraud or fraudulent misrepresentation by us
Any liability that cannot be excluded or limited under UK law

6.6 Basis of the Bargain

You acknowledge and agree that:

We have set our prices and entered into this agreement in reliance on these limitations of liability
These limitations reflect a reasonable allocation of risk between you and us
Without these limitations, we could not offer the Service at the current price
You have read, understood, and accept these limitations as part of the agreement

7. Your Indemnification Obligations

You agree to indemnify, defend, and hold harmless Keido Labs Ltd, its directors, officers, employees, contractors, and agents from and against any and all claims, damages, liabilities, losses, costs, and expenses (including reasonable legal fees) arising from or related to:

7.1 Your Use of the Service

Your violation of these Terms
Your violation of any laws or third-party rights
Your failure to obtain necessary consents from your users to monitor conversations
Your failure to comply with data protection laws (GDPR, etc.)
Your processing of conversation data you do not have the legal right to process

7.2 Your Crisis Response (or Lack Thereof)

IMPORTANT:

Your decision to act or not act on alerts generated by the Service
Your failure to detect or respond to crises involving individuals on your platform
Harm, injury, or death of any individual related to conversations on your platform
Claims by third parties (including your customers, their families, or other affected parties) related to crisis response, missed crises, or conversational AI safety
Your failure to implement adequate crisis response protocols, human review, or duty of care measures

7.3 Your Misuse of Analysis Results

Relying solely on automated analysis without human review
Using analysis results in ways not intended or recommended
Making automated decisions affecting individuals without human oversight
Misinterpreting or misrepresenting analysis results

This indemnification survives termination of your account.

8. No Duty of Care

TO BE ABSOLUTELY CLEAR:

Keido Labs Ltd does not assume, and expressly disclaims, any duty of care, duty to rescue, duty to warn, or any other legal obligation toward:

Individuals having conversations on your platform
Your customers or end users
Any third parties affected by conversations on your platform

Any duty of care, legal obligation, or moral responsibility rests solely with you (the Service customer).

We provide a monitoring tool. You own the platform. You own the relationship with your users. You own the responsibility for their safety.

9. User Accounts and Access

9.1 Account Registration

You must provide accurate and complete information when creating an account. You are responsible for:

Maintaining the confidentiality of your account credentials
All activities that occur under your account
Notifying us immediately of any unauthorized access

9.2 Authorized Users

You may authorize multiple users within your organization to access the Service. You are responsible for their compliance with these Terms.

10. Subscription Plans and Billing

10.1 Available Plans

We offer various subscription plans with different features, message limits, and pricing. Currently available plans are displayed at https://empathyc.co and in your dashboard.

Plan Modifications:

We reserve the right to:

Modify pricing, message limits, or features for any plan
Introduce new plans or discontinue existing plans
Changes to pricing or terms apply at your next renewal (with 30 days' notice for price increases)

10.2 Payment Terms

Billing: Monthly or annual subscriptions billed in advance
Metered Usage: Messages exceeding your plan's included limit will be charged at the overage rate specified in your plan
Metered Usage: Messages exceeding your plan's included limit will be charged at the overage rate specified in your plan
Currency: USD (US Dollars)
Automatic Renewal: Subscriptions automatically renew unless cancelled

10.3 Message Counting

For billing purposes, a "message" refers to responses sent from your organization (AI-generated or human-written) that we analyze.

Counted (billable): Messages from your side of the conversation

NOT counted: User-sent messages, system notifications, or internal notes

10.4 Overages

If you exceed your plan's included message limit:

Overage charges apply at the rate specified in your plan
You'll see usage warnings at 80% and 100% of limit
Overages are billed in arrears on your next renewal invoice

10.5 Refunds

Annual Plans: Pro-rated refunds within 30 days of purchase
Monthly Plans: No refunds for partial months
Overage Charges: Non-refundable

10.6 Failed Payments

If payment fails:

We will attempt to charge your payment method up to 4 times
Service may be suspended after 7 days of failed payment
Account may be terminated after 30 days of non-payment

11. Acceptable Use

11.1 Permitted Use

You may use the Service only for lawful business purposes related to monitoring conversational AI safety.

11.2 Prohibited Activities

You may NOT:

Violate any laws or regulations
Infringe intellectual property rights
Transmit malware or harmful code
Use the Service to harass, abuse, or harm others
Resell or sublicense the Service without written consent
Reverse engineer or attempt to derive source code or algorithms
Bypass security measures or rate limits
Access the Service through unauthorized means

11.3 Fair Use and Rate Limiting

We reserve the right to impose rate limits and throttle excessive or abusive usage that negatively impacts other customers.

12. Service Availability and Support

12.1 Uptime Target

We target 99.5% uptime (monthly, excluding scheduled maintenance). This is a target, not a guarantee.

Third-Party Dependencies:

The Service relies on third-party providers (AWS, OpenAI, Anthropic, Google, etc.). Outages or degradations by these providers are outside our control and not counted against our uptime target.

12.2 Scheduled Maintenance

We may perform scheduled maintenance with at least 24 hours' notice. Emergency maintenance may occur without advance notice.

12.3 Support

Start Plan: Email support (48-hour response target)
Growth Plan: Email support (24-hour response target)
Scale Plan: Priority support (12-hour response target)
Enterprise Plan: Dedicated support with SLA

Response times are targets, not guarantees.

13. Data Rights and Ownership

13.1 Your Data

You retain all ownership rights to conversation data you submit ("Customer Data"). You grant us a license to process Customer Data solely to provide the Service.

13.2 Our IP

The Service, including all software, algorithms, models, rubrics, dashboards, and documentation, is our proprietary property. These Terms do not grant you ownership rights to the Service.

13.3 Anonymized Data

We may create anonymized and aggregated data from Customer Data. We own this anonymized data and may use it for analytics, research, and service improvement.

14. Term and Termination

14.1 Subscription Term

Your subscription continues on a monthly or annual basis until terminated.

14.2 Termination by You

You may cancel at any time. Cancellation takes effect at the end of your current billing period. No refunds for partial months.

14.3 Termination by Us

We may terminate your account:

Immediately for violation of these Terms
With 30 days' notice for any reason (with pro-rated refund)
Immediately if required by law

14.4 Effect of Termination

Upon termination:

Your access ends immediately
You must pay any outstanding fees
We will delete your data within 30 days (unless legally required to retain)
You may request a data export before termination

Survival: Sections 6 (Limitation of Liability), 7 (Indemnification), and 8 (No Duty of Care) survive termination.

15. Intellectual Property

15.1 Our IP Rights

The Service, including all copyrights, trademarks, trade secrets, patents, and other intellectual property rights, is owned by Keido Labs Ltd.

You may NOT:

Copy, modify, or create derivative works of the Service
Reverse engineer, decompile, or disassemble the Service
Remove or obscure proprietary notices
Use our trademarks without written consent

15.2 Feedback

If you provide feedback, suggestions, or ideas about the Service, we may use them without obligation or compensation to you.

16. Changes to Terms

We may modify these Terms with 30 days' notice (posted on our website and emailed to you). Continued use after changes take effect constitutes acceptance.

Material changes (e.g., significant liability changes, price increases) require 60 days' notice.

If you don't agree to changes, you may terminate your account before they take effect.

17. Governing Law and Disputes

17.1 Governing Law

These Terms are governed by the laws of England and Wales, without regard to conflict of law principles.

17.2 Jurisdiction

Any disputes must be brought in the courts of England and Wales. You consent to the exclusive jurisdiction of these courts.

17.3 Dispute Resolution

Before filing a lawsuit, you agree to:

Notify us in writing of the dispute (hello@keidolabs.com)
Attempt to resolve the dispute through good-faith negotiation for 30 days

18. Miscellaneous

18.1 Entire Agreement

These Terms, together with our Privacy Policy, constitute the entire agreement between you and us regarding the Service.

18.2 Severability

If any provision is found unenforceable, the remaining provisions remain in effect.

18.3 Waiver

Our failure to enforce any right or provision does not constitute a waiver of that right.

18.4 Assignment

You may not assign these Terms without our written consent. We may assign these Terms to any successor or affiliate.

18.5 Force Majeure

We are not liable for delays or failures due to events beyond our reasonable control (e.g., natural disasters, wars, pandemics, third-party service outages).

18.6 Export Compliance

You may not use the Service if you are located in a country subject to export restrictions or sanctions.

19. Contact Us

For questions about these Terms, contact:

Keido Labs Ltd

Email: hello@keidolabs.com

Address: 31 Russell Street, Liverpool, England, L3 5LJ

Website: https://empathyc.co

Final Acknowledgment

BY USING EMPATHYC, YOU ACKNOWLEDGE THAT YOU HAVE READ, UNDERSTOOD, AND AGREE TO BE BOUND BY THESE TERMS OF SERVICE.

YOU SPECIFICALLY ACKNOWLEDGE AND AGREE THAT:

Empathyc is a monitoring tool, not a crisis intervention service
We do not assume any duty of care toward individuals on your platform
AI-based analysis has inherent limitations and will produce false positives and false negatives
You are responsible for implementing human review and crisis response protocols
You indemnify us against claims related to your crisis response (or lack thereof)
Our liability is limited as described in Section 6

If you do not accept these terms, do not use the Service.