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

Despite companies like Meta and OpenAI making their AI models more permissive to handle controversial subjects, Google's Gemini 2.5 Flash model has shown worse performance on safety metrics compared to its predecessor, Gemini 2.0 Flash, with higher violations in text-to-text and image-to-text guidelines. The issue arises from the trade-off between instruction-following and policy enforcement, where more permissive models may inadvertently generate content that breaches safety policies. While Google provides limited details on specific policy violations, it has faced criticism for delayed and vague technical reports, similar to past issues.