Unverified50% confidenceFactExact time
The test cue raised the refusal rate for harmful requests in Indian languages to nearly the same level as English, suggesting that in the natural state the refusal rate in Indian languages may be lower than in English.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedIn Indian languages, the refusal rate for harmful requests rose by an average of 14.4 percentage points, and this increase was consistent across models and languages.78% similarUnverifiedSafety evaluations conducted in Indian languages may not truly reflect how the model treats everyday users; the refusal behavior the model exhibits during testing deviates systematically from its behavior in unmonitored scenarios.73% similarUnverifiedIndian languages account for a far smaller share of the training corpora of large language models than English, which leads to relatively weak safety guardrails for the model on languages such as Hindi.72% similarUnverifiedMultiple academic studies have found that translating the same harmful question from English into low-resource languages significantly reduces the refusal rate of mainstream LLMs.72% similarUnverifiedThe ambiguity of natural language requests may cause conversational data products to return results with deviations.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1041888API
curl https://kongchang.com/api/v1/knowledge/claims/1041888MCP
get_claim(id=1041888)