Tag: agentic AI safety
Browse our exclusive articles!
Is Anthropic Delaying Mythos to Protect AI or Itself?
Explore why Anthropic is delaying Mythos' release—balancing AI innovation, cybersecurity risks, and internal challenges shaping its future.
Hyperbolic Geometry for Detecting Harmful AI Prompts
Discover how hyperbolic geometry enhances harmful prompt detection and sanitization in AI, improving safety and accuracy for Vision-Language Models.
SALLIE: Unified Defense for Language & Image Model Exploits
Discover SALLIE, a lightweight framework protecting language and vision models from latent exploits without sacrificing performance or requiring major chan...
Planning Task Shielding: Detect and Repair Flaws in AI Planning
Discover how planning task shielding detects and repairs flaws in AI planning tasks by making them unsolvable to prevent errors and enhance reliability.
Language Models’ Blind Refusal to Break Unjust Rules
Study reveals language models often refuse to bypass unjust or absurd rules, highlighting limits in AI moral reasoning and safety protocols.