AI Ethics & Responsible Use
Last Updated: December 17, 2025
Our Commitment
Our AI Principles
We clearly communicate how our AI works, its limitations, and potential risks.
We take responsibility for our AI infrastructure, security, and user experience.
We prioritize user safety, data security, and minimizing harm over growth.
We give users control over their data, AI provider selection, and usage.
What We Take Responsibility For
- Data Security:We ensure your documents are encrypted, securely stored, and protected from unauthorized access.
- Service Availability:We maintain reliable infrastructure and uptime for document access and AI features.
- Privacy Protection:We implement row-level security, comply with GDPR, and never sell your data.
- Credit Accuracy:We accurately track and charge credits based on actual AI token usage.
- Document Processing:We ensure documents are properly extracted, chunked, and indexed for accurate retrieval.
- Context Selection:We use smart algorithms (hybrid search, query expansion) to provide relevant context to AI.
- Source Attribution:We provide citations showing which documents and pages AI responses are based on.
- Provider Selection:We vet AI providers for reliability, safety, and ethical practices before integration.
Known Limitations & What Can Go Wrong
AI is powerful but not perfect. Here are known issues you should be aware of:
- Hallucinations (False Information):AI can generate confident-sounding but incorrect information, especially for questions not directly answerable from your documents.
Mitigation: Always verify critical information from source documents. We provide citations for this reason.
- Context Limitations:AI only sees ~25 document chunks per query (~20,000 tokens), not your entire document library.
Mitigation: Ask specific questions. For broad queries, ask multiple follow-up questions or use document summaries first.
- OCR Errors:Scanned PDFs with poor quality may have text extraction errors, leading to incorrect information.
Mitigation: Upload high-quality, digital PDFs when possible. Review OCR results before relying on them.
- Outdated Knowledge:AI models have knowledge cutoffs (e.g., GPT-4 trained on data up to Oct 2023). They don't know recent events outside your documents.
Mitigation: Use Perplexity Pro for real-time web information, or upload recent documents.
- Bias:AI models trained on internet data may exhibit cultural, gender, racial, or other biases.
Mitigation: Use multiple AI providers, critically evaluate responses, and report biased outputs.
- Misinterpretation:AI may misunderstand ambiguous queries, technical jargon, or complex document structures.
Mitigation: Rephrase your question, provide more context, or specify document sections explicitly.
- Calculation Errors:While DeepSeek and others are good at math, complex calculations may have errors.
Mitigation: Verify all calculations independently, especially for financial or engineering decisions.
- Language Limitations:AI performs best in English. Non-English content may have lower accuracy.
Mitigation: Use English for critical queries. For multilingual documents, specify language in your question.
What We Don't Take Responsibility For
- Decisions Based on AI Outputs:You are responsible for verifying AI responses and making informed decisions. We provide tools, not professional advice.
- Legal, Financial, or Professional Advice:AI responses are informational only. Consult licensed professionals for legal, financial, engineering, or medical advice.
- Third-Party AI Provider Errors:We use Google, OpenAI, Anthropic, DeepSeek, xAI, and Perplexity. While we select reputable providers, we cannot control their model outputs or policies.
- Accuracy of User-Uploaded Documents:We process documents as-is. We are not responsible for errors, outdated info, or misleading content in your uploaded files.
- Misuse of the Service:If you use the Service for illegal purposes, generate harmful content, or violate terms, you bear full responsibility.
- Business Losses:We are not liable for financial losses, missed opportunities, or business decisions made based on AI outputs.
- Internet Connectivity:The Service requires internet access. We are not responsible for your network issues or ISP outages.
Safe Use Guidelines
✅ Always Verify Critical Information
For contract terms, financial figures, dates, specifications, or anything legally binding, verify directly from source documents.
✅ Use Source Citations
Click on source citations to review original document pages. Don't rely solely on AI summaries for important decisions.
✅ Ask Follow-Up Questions
If an answer seems incomplete or unclear, ask for clarification or request specific document references.
✅ Choose the Right AI Provider
Use Gemini for speed, Claude for long documents, GPT-4 for structured outputs, DeepSeek for reasoning, Perplexity for web search.
✅ Provide Clear Context
Specify document names, sections, or time periods in your queries for more accurate responses.
✅ Report Issues
If you encounter inaccurate, biased, or inappropriate AI responses, report them so we can improve.
❌ Don't Rely Solely on AI for Legal/Financial Decisions
Always consult qualified professionals for legal contracts, financial planning, or engineering designs.
❌ Don't Upload Sensitive Personal Data
Avoid uploading documents with social security numbers, credit card info, or medical records unless absolutely necessary.
❌ Don't Assume AI Has All Context
AI only sees selected document chunks, not your entire project. It may miss important information from other files.
AI Provider Transparency
We use 6 third-party AI providers. Here's what you should know about each:
Strengths: Fast, cost-effective, excellent for PDFs and images, large context window (1M tokens)
Limitations: May occasionally produce generic responses, less "thoughtful" than Claude
Data Handling: Google may use queries for model improvement unless you opt-out. See Google AI Terms.
Strengths: Extended reasoning, excellent for calculations and technical analysis, very affordable
Limitations: Slower response times, may over-explain simple queries
Data Handling: DeepSeek processes queries but claims not to train on user data. See DeepSeek Privacy Policy.
Strengths: Safety-focused, excellent for contracts, 200K context window, nuanced understanding
Limitations: Higher cost, can be overly cautious, may refuse benign queries
Data Handling: Anthropic does not train on user data. Strong privacy commitment.
Strengths: Function calling, structured outputs, reliable performance, latest knowledge
Limitations: Higher cost, knowledge cutoff (Oct 2023), occasional verbosity
Data Handling: OpenAI may use API data for abuse monitoring but not training (as of March 2023).
Strengths: Up-to-date information, reasoning mode, fast inference
Limitations: Less tested, may have less refined outputs, limited context window
Data Handling: xAI data handling policies are evolving. See xAI Terms.
Strengths: Real-time web search, citation-backed responses, external sources
Limitations: Higher cost, slower, may include irrelevant web results
Data Handling: Perplexity processes queries with web search. See Perplexity Privacy Policy.
Bias & Fairness
Potential for Bias
What We're Doing About It:
- • Using multiple AI providers to balance different perspectives
- • Implementing prompt engineering to reduce harmful outputs
- • Monitoring AI responses for problematic content
- • Allowing users to report biased or inappropriate responses
- • Regularly updating our AI configurations based on feedback
Your Role: If you encounter biased or inappropriate AI responses, please report them to ethics@insights.aecos.app
Environmental Impact
AI models require significant computational resources. We're committed to minimizing environmental impact:
- • Using efficient models like DeepSeek and Gemini 2.0 Flash (optimized for cost and energy)
- • Caching responses to avoid redundant AI calls
- • Smart context selection to minimize token usage
- • Partnering with providers using renewable energy (Google, Anthropic)
- • Providing transparency on credit usage so users can make informed choices
Reporting Issues
Report AI Issues:
- • Inaccurate Information: ai-accuracy@insights.aecos.app
- • Biased or Harmful Content: ethics@insights.aecos.app
- • Privacy Concerns: privacy@insights.aecos.app
- • General Feedback: feedback@insights.aecos.app
When Reporting, Include:
- • Your query/prompt
- • AI provider used (Gemini, Claude, GPT-4, etc.)
- • Screenshot or copy of the response
- • What went wrong and expected behavior
Governance & Updates
We continuously monitor and update our AI practices:
- • Quarterly AI Ethics Reviews: Internal audits of AI outputs and user feedback
- • Provider Updates: We update to the latest, safest AI models
- • Transparency Reports: Annual reports on AI usage, issues, and improvements
- • User Feedback Loop: All reported issues are reviewed within 7 days
- • Policy Updates: This page is updated whenever we make significant changes
Contact Us
Email: ethics@insights.aecos.app
Phone: +91 98191 23672
Address: Plot 18A, Flat B902, Media Society, Dwarka 1, South Delhi 110075, India
This AI Ethics Policy is governed by the laws of India and subject to the exclusive jurisdiction of the courts located in New Delhi, India.