Tag: agentic AI safety
Browse our exclusive articles!
Human-Guided Harm Recovery for Safer AI Agents
Explore human-guided harm recovery methods that help AI agents safely recover from harmful actions, ensuring responsible computer system use.
ARES: Enhancing AI Safety with Adaptive Red-Teaming
Discover how ARES improves AI safety by adaptive red-teaming and repairing policy-reward systems in large language models.
Robust Policy Learning Against Adversaries with Regret Guarantees
Discover robust policy learning methods under adversarial conditions with regret and violation guarantees for safer AI decision-making systems.
Jailbreak Scaling Laws in Large Language Models Explained
Explore how adversarial attacks cause polynomial to exponential growth in jailbreak success rates of large language models, impacting AI safety.
Red-Teaming Web-Augmented Large Language Models Safely
Explore CREST-Search, a red-teaming framework to identify and mitigate risks in web-augmented large language models for safer AI search results.