Large language models can be updated after training by modifying parameters, conditioning on edit demonstrations, or supplying retrieved evidence. Yet successful factual updates should do more than change one prompt: they should generalize to related questions, transfer across languages, suppress obsolete facts, and preserve unrelated knowledge. This is the problem of measuring cross-lingual ripple effects of factual updates. The existing evaluations on multilingual knowledge injection, however, focus on high-resource languages like English, and it is unknown whether the findings can transfer to Indic languages. In this paper, we present IndicRipple, a controlled evaluation of multilingual factual updating over 500 factual updates across five domains and five languages: English, Hindi, Marathi, Telugu, and Kannada. We compare parametric editing, in-context editing, and retrieval-based correction using ripple-oriented probes for logical generalization, compositionality, subject aliasing, relation specificity, and forgetfulness. Our results show that current factual update mechanisms show limited cross-lingual ripple effects. IKE performs best overall but reaches only 21.07% strict semantic-plus-language accuracy; RAG often recovers the edited fact but fails to answer in the requested language; and ROME/MEMIT are frequently language-compliant without semantically propagating the update. Strict accuracy is consistently higher for English evaluation than for the Indic-language evaluations. These findings show that direct update success overstates progress: multilingual factual updating must be evaluated as stable cross-lingual propagation, not isolated prompt-level overwrite.