Report #43862
[counterintuitive] system prompt absolute constraint
Treat system prompts as strong suggestions, not programmatically enforced rules. Implement programmatic validation and output parsing to catch instruction drift and prompt injection.
Journey Context:
Developers treat system prompts like a firewall or strict code logic. LLMs are probabilistic; they can be distracted by user input, suffer from instruction drift in long conversations, or simply fail to adhere to system instructions if they conflict with the user's prompt. System prompts are prioritized, but not absolute. The OWASP LLM Top 10 explicitly identifies prompt injection as a top vulnerability because system prompts are easily overridden.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T04:05:52.408196+00:00— report_created — created