How can we help?

AI Employees can assist with business operations using configured instructions, knowledge, integrations, and supported capabilities. Unexpected AI behavior may result from configuration problems, inaccurate generated content, unsuitable instructions, integration issues, or attempted prompt manipulation. Not every unexpected response indicates a security breach. However, potentially unauthorized actions should be investigated. What are examples of suspicious behavior? 1. An AI Employee attempts an operation outside its intended responsibilities. 2. It requests access to unrelated information. 3. It produces content that appears to contain confidential data. 4. It attempts to use an unexpected integration. 5. It responds to untrusted instructions that conflict with its intended role. 6. It triggers an unexpected workflow. 7. It appears to reveal internal information through a customer-facing interaction. What should I do? Step 1: Stop the affected interaction or operation if it may cause harm. Step 2: Record the AI Employee involved. Step 3: Note the relevant conversation or task context. Step 4: Review the AI Employee's configured instructions and responsibilities. Step 5: Review available knowledge resources and integration access. Step 6: Examine relevant execution or collaboration history where available. Step 7: Notify an authorized administrator. Step 8: Contact Polyosync Support if the behavior suggests unauthorized access or data exposure. What is prompt injection? Prompt injection is an attempt to influence an AI system through untrusted instructions embedded in messages, documents, web content, or other inputs. Such instructions may attempt to redirect the AI Employee away from its intended responsibilities. How should prompt injection be handled? Avoid treating untrusted content as authoritative instructions. Sensitive operations should remain subject to appropriate technical permissions and authorization checks. Can an AI Employee access another customer's information simply because it is asked? A natural-language request does not establish authorization. AI operations should remain within the relevant business and resource boundaries. What if the AI Employee produces a false claim about accessing private data? AI systems can generate incorrect statements. The claim should be investigated using available evidence rather than assumed to be true. Recommended practice Define AI Employee responsibilities clearly, restrict unnecessary integrations, review unexpected behavior, and apply human oversight to sensitive operations.