Bloomberg study finds RAG systems increase unsafe responses in Llama-3-8B from 0.3% to 9.2%, raising safety concerns for LLMs; suggests using domain-specific AI safety taxonomies in financial services to address risks

RAG systems risk undermining LLM safety; Bloomberg study reports a jump in unsafe responses, from 0.3% to 9.2% with Llama-3-8B, when using RAG. Bloomberg suggests domain-specific AI safety taxonomies for financial services to mitigate risks.

Source: venturebeat.com