AI & Data Scraping Ethics Policy

At Soft Clerk, we architect cutting-edge AI chatbots, workflow agents, and semantic intelligence systems with safety, predictability, and business transparency at their core.

We adhere to strict engineering guardrails to prevent harmful model behaviors, protect enterprise data privacy, and deliver transparent AI applications that augment human productivity.

Zero Model Training on Client Data: We configure our enterprise LLM API integrations (OpenAI, Anthropic, Google Gemini, OpenRouter) with strict zero-data-retention (ZDR) and non-training flags.

Your proprietary source code, internal knowledge bases, and customer messages are never utilized to train foundation models or shared across multi-tenant boundaries.

To ensure reliable, deterministic outputs in business-critical software, our AI engineering practices incorporate:

• Retrieval-Augmented Generation (RAG) with vector verification to ground model responses in validated facts.
• Schema validation with Zod to enforce strict structured JSON outputs from LLM function calls.
• Prompt injection defenses and input sanitization layers to protect autonomous agents from adversarial manipulation.
• Human-in-the-loop (HITL) approval gates for sensitive operations like financial transactions or mass email dispatches.

Web data extraction is a foundational capability for market intelligence, price monitoring, and research. We conduct all web crawling according to established ethical industry benchmarks:

• Public Data Focus: We only extract publicly accessible information and do not harvest data behind paywalls or authenticated private intranets without explicit client authorization.
• Polite Crawling Cadence: We implement intelligent rate-limiting, concurrency caps, and request throttling to avoid placing undue load on target servers.
• Respect for Privacy: We implement automated scrubbing filters to eliminate sensitive PII (Personally Identifiable Information) from raw crawl feeds before pipeline delivery.

Soft Clerk strictly refuses to build software, AI agents, or data scraping clusters designed for:

• Spam generation, phishing, or deceptive impersonation.
• Circumventing copyright protection systems or extracting proprietary software binaries.
• Surveillance, unauthorized tracking of individuals, or gathering discriminatory profiling datasets.
• Denial of service (DoS) or deliberate disruption of web services.