#Harmful

A new arXiv paper identifies the mechanism: harmless, narrow fine-tuning can induce broad, unexpected misalignment in large language models via feature superpos

Targeted tweaks to specific attention heads can slash jailbreak success rates by several‑fold (e.g.,...

From Independent RSS Feed: Government commits to new social media restrictions for under-16s in ‘massive step forward’

Google DeepMind releases new findings and an evaluation framework to measure AI's potential for harmful manipulation in areas like finance and health, with the goal of enhancing AI safety.