Tag: agentic AI safety
Browse our exclusive articles!
Gemma Scope 2: Advanced AI Safety for Language Models
Discover how Gemma Scope 2 enhances AI safety by providing tools to analyze and understand complex language model behavior effectively.
Preventing Harmful AI Manipulation: DeepMind’s Approach
Discover how Google DeepMind combats harmful AI manipulation in finance, healthcare, and society with innovative safety measures and education.
Finetuning Triggers Verbatim Recall in Large Language Models
Study reveals finetuning LLMs can cause verbatim recall of copyrighted books, exposing major copyright and safety vulnerabilities in AI models.
Boosting Jailbreak Attacks on LLMs Using Persona Prompts
Explore how persona prompts enhance jailbreak attacks on LLMs, revealing vulnerabilities and impacting AI safety with new genetic algorithm methods.
Relationship-Aware Safety Unlearning for Safer Multimodal LLMs
Discover relationship-aware safety unlearning to prevent unsafe associations in multimodal LLMs while preserving benign concept uses effectively.