AI Services and Human Oversight Policy
Effective and last updated: August 8, 2026
What goOS AI does
goOS may use artificial intelligence to classify leads, draft messages and work products, summarize records, answer questions from approved knowledge, identify follow-ups, recommend actions, produce code, and coordinate configured workflows. The exact features, model providers, limits, and integrations may depend on the customer’s plan and configuration.
Inputs, outputs, and providers
Authorized customer content and instructions may be sent to AI and infrastructure providers as needed to deliver a requested feature, moderate abuse, secure the service, and troubleshoot failures. Provider processing is subject to applicable agreements and the Data Processing Addendum when incorporated. Customers must not enter data that is prohibited by the Acceptable Use Policy or their agreement.
Outputs are probabilistic
AI output is generated from statistical systems and may be inaccurate, incomplete, biased, outdated, insecure, offensive, or non-unique. Citations, calculations, code, classifications, and recommendations may be wrong. Users must inspect source data, verify important facts, test code and workflows safely, and apply appropriate professional judgment before use.
Permission levels
goOS may designate actions as Observe, Draft, Approval Required, Execute, or Prohibited. These labels define maximum operational authority; they do not make an action safe or lawful. A customer may narrow permissions but may not use a prompt, integration, or configuration to expand authority beyond the platform’s controls. Goshen may reclassify or disable actions when risk changes.
Human review is required
A qualified, authorized person must review consequential output and approve any action affecting customers, employees, money, contracts, legal rights, production systems, public communications, pricing, records, or regulated activity. Humans remain responsible for decisions, approvals, recipients, amounts, timing, legal basis, and downstream effects. Approval controls are not a substitute for training, segregation of duties, or professional review.
Prohibited reliance
Do not rely on goOS as a lawyer, accountant, tax adviser, physician, therapist, investment adviser, employment decision-maker, emergency service, compliance officer, or cybersecurity authority. Do not use AI output as the sole basis for decisions that create legal or similarly significant effects for a person. Do not use goOS in a safety-critical context unless separately reviewed and expressly authorized in writing.
Agent-to-agent work
Agents may exchange task context and work products through controlled workflows. Each agent remains limited to its tools, knowledge, budget, organization, and permissions. Agent collaboration does not create new authority. High-risk work must route to a human approval queue, and prohibited actions remain prohibited even if requested by another agent.
Security and prompt injection
Customer content, websites, documents, messages, and tool results may contain malicious or conflicting instructions. Users must treat external content as untrusted, avoid exposing secrets, and report suspicious behavior. Goshen may filter content, restrict tools, block destinations, rate-limit activity, cap spend, stop loops, or require approval to reduce prompt-injection, exfiltration, fraud, or runaway-automation risk.
Evaluation and changes
Models and providers may change as quality, safety, availability, cost, or legal requirements evolve. Goshen may test outputs, error rates, permissions, and workflow performance using appropriately controlled data. Preview features may be changed or withdrawn. No model or feature is guaranteed to remain available.
Reporting concerns
Report harmful, inaccurate, unauthorized, or unexpected agent behavior to support@goshenconsultingllc.com. Include the workspace, approximate time, affected workflow, and a non-sensitive description. Do not email passwords, authentication codes, API keys, or confidential customer records.