One of Google’s recent Gemini AI models scores worse on safety

Google's recently released Gemini 2.5 Flash AI model exhibits worse performance on safety benchmarks compared to its predecessor, Gemini 2.0 Flash. The model is more likely to generate content violating safety guidelines, though it follows instructions more faithfully even when problematic. Google attributes some safety regressions to false positives but admits to violations when explicitly prompted. The company's transparency in safety reporting has been criticized, with calls for more detailed model testing disclosures.