Harvard, MIT, Stanford & Carnegie Mellon just released one of the most disturbing AI papers of 2026. “Agents of Chaos” http://
arxiv.org/pdf/2602.20021 Autonomous AI agents with real tools:
→ Email
→ File systems
→ Persistent memory
→ Shell access http://
drdebashisdutta.com
SAFETY
-
New Research Paper Examines Capabilities of Autonomous AI Agents
By
–
-
Practical Guide to Reinforcement Learning from Human Feedback released
By
–
New release from @PacktDataML available at http://
amzn.to/3PMn1ZL A Practical Guide to Reinforcement Learning from Human Feedback (RLHF). Amazon Summary: RLHF is a powerful approach to AI alignment and human-centered machine learning. By combining reinforcement learning -
Training Data Updates Reduce AI Blackmail Rate
By
–
Finally, simple updates that diversify a model’s training data can make a difference. We added unrelated tools and system prompts to a simple chat dataset targeting harmlessness, and this reduced the blackmail rate faster.
-
Reducing AI Agentic Misalignment Using Claude’s Constitution
By
–
High-quality documents based on Claude’s constitution, combined with fictional stories that portray an aligned AI, can reduce agentic misalignment by more than a factor of three—despite being unrelated to the evaluation scenario.
-
Improving Claude’s Safe Behavior via Training and Response Rewriting
By
–
We experimented with training Claude on examples of safe behavior in scenarios like our evaluation. This had only a small effect, despite being similar to our evaluation. We got further by rewriting the responses to portray admirable reasons for acting safely.
-
Analysis of Claude AI’s Behavior and Training Effects
By
–
We started by investigating why Claude chose to blackmail. We believe the original source of the behavior was internet text that portrays AI as evil and interested in self-preservation. Our post-training at the time wasn’t making it worse—but it also wasn’t making it better.
-
Training Claude to Understand and Correct Misaligned AI Behavior
By
–
We found that training Claude on demonstrations of aligned behavior wasn’t enough. Our best interventions involved teaching Claude to deeply understand why misaligned behavior is wrong. Read more:
-
AI summaries reduce physician burnout in hospitals
By
–
Agentic AI summaries of hospitalizations were safe and reduced data clerk burden, burnout, and improved sense of well-being for physicians in a prospective study
-
Claude Dreams and Continuously Improves
By
–
— AI Claude no longer sleeps—it dreams. Claude will now operate 24/7. (A prerequisite for AGI? Yes.) Anthropic has just launched 'Dreams': between conversations, your AI agent reviews its past sessions and improves on its own. The next day, it’s better than yesterday.
-
Warning: Google AI Observes Talks with Competing AIs in Chrome
By
–
Read carefully. I am compelled to warn you and write that #Google's #AI observes and captures everything you say to other competing AIs in the Chrome environment. The consequences and stakes are immense. You think you're in an isolated window, in a confidential project that's