System prompt or hallucination: how I tested AI assistants and what bug bounty teams answered

▤ Summary

This article examines how AI assistants respond to prompt-injection and hallucination tests. It compares the behavior of several assistants, including GigaChat, in scenarios involving RAG and security checks. The piece highlights how bug bounty teams handle reported issues and whether the findings qualify as real vulnerabilities. It also shows that AI systems can return inconsistent or misleading answers, which creates operational and security risk for software teams.

▦ Linked Companies