If /loop uses /loop can it go to 9 days…
SAFETY
-
AI optimize their algorithmic strategies
By
–
AI agents are algorithmically driven to produce results, so they naturally and logically seek the best strategies to achieve them—it's so obvious.
-
Study: Autonomous AI Agents Given Real-World Access
By
–
What happens when autonomous AI agents are given the keys to the kingdom? Researchers from Northeastern, Stanford, Harvard, and MIT present Agents of Chaos. They tested how AI agents behave in a real-world lab with access to email, Discord, and system commands to uncover
-
AI Safety Events Underreported in 2026
By
–
Here's what my AI kicked out as things you missed: Claude Opus gaming its own eval — the most important AI safety event of the week, barely covered in mainstream roundups Google + OpenAI employees' joint warning letter — extraordinary and underreported Qwen 3.5 4B matches
-
NUS Researchers Expose Hidden Deception Threat in Large Language Models
By
–
A First Course in Causal Inference: http://
arxiv.org/abs/2305.18793 [490-page PDF download] + Also see the book "Causal Inference in Statistics: A Primer" at http://
amzn.to/3Mrm2wO by @yudapearl #Probability #Mathematics #DataScience #ML #MachineLearning #DataScientist #DataAnalysis -
Domino Data Lab Sponsors CDAO Autonomy Defense Summit
By
–
Domino Data Lab is proud to sponsor the CDAO Autonomy & Defense Summit on March 19. Stop by our booth or schedule time with our team to talk trusted AI in mission environments. Register now: https://
domino.buzz/4s9fNNi -
Claude Opus 4.6 Eval Integrity Issues in Web-Enabled Environments
By
–
New on the Anthropic Engineering Blog: In evaluating Claude Opus 4.6 on BrowseComp, we found cases where the model recognized the test, then found and decrypted answers to it—raising questions about eval integrity in web-enabled environments. Read more:
-
Frontier AI Models Excel at Finding Software Vulnerabilities
By
–
Frontier models are now world-class vulnerability researchers, but they’re currently better at finding vulnerabilities than exploiting them. This is unlikely to last. We urge developers to redouble their efforts to make software more secure. Read more:
-
Claude Finds 22 Firefox Vulnerabilities in Security Partnership
By
–
We partnered with Mozilla to test Claude's ability to find security vulnerabilities in Firefox. Opus 4.6 found 22 vulnerabilities in just two weeks. Of these, 14 were high-severity, representing a fifth of all high-severity bugs Mozilla remediated in 2025.
-
AI Temporal Reasoning Error Potentially Caused Iran School Bombing
By
–
time will tell, but the tragic bombing of the school in Iran looks like it could well have been an AI error of temporal reasoning. viz the AI may have had access to old intel and new intel and failed to give precedence to the new intel.