Open-source safety method judges LLM requests by degree, not yes/no
I-Lang v5.0 is an protocol for judging LLM safety without reducing every request to allow or deny. Common s compare requests against fixed rules and patterns, then make a yes-or-no decision. Stricter rules can block valid research, medical questions, or security analysis, while looser rules can miss clever attempts.
I-Lang v5.0 evaluates each request across 9 areas: intent, capability, consequence, relationship, certainty, authority, reversibility, evidence, and sovereignty. The output is meant to choose the most useful cooperative action for the situation, instead of only blocking or allowing. Its design says no rule should have absolute power; each rule gets a weight between 0 and 1.
It is released under the , so other projects can reuse or adapt it with few restrictions.
Key points
- I-Lang v5.0 replaces simple allow/deny s with graded judgment.
- It scores requests across 9 areas, including intent, consequence, authority, and evidence.
- The goal is to reduce both overblocking valid requests and missing risky ones.
- The system chooses a suitable action, not just a yes-or-no answer.
- The makes it easier for agent builders to test or adapt it.