This research was led by Jackson Kaunismaa through the MATS program and supervised by researchers at Anthropic, with additional support from Surge AI and Scale AI. Read the full paper: https://
arxiv.org/pdf/2601.13528
SAFETY
-
Anthropic-Led Research on AI Model Improvement Published
By
–
-
Frontier Model Training Enables More Capable Dangerous Open-Source AI
By
–
These attacks scale with frontier model capabilities. Across both OpenAI and Anthropic model families, training on data from newer frontier models produces more capable—and more dangerous—open-source models.
-
AI Adolescence: National Security, Economy and Democracy Risks
By
–
The Adolescence of Technology: an essay on the risks posed by powerful AI to national security, economies and democracy—and how we can defend against them:
-
Machines of Loving Grace: AI potential when implemented correctly
By
–
It's a companion to Machines of Loving Grace, an essay I wrote over a year ago, which focused on what powerful AI could achieve if we get it right:
-
Databricks AI Guardrails Block Sensitive Data Locally
By
–
[DEMO] Databricks AI Guardrails block sensitive data before it ever leaves your environment.@dennylee shows how PII like UK National Insurance numbers and credit card data is detected and stopped locally at the serving endpoint, instead of being sent to an external model.… pic.twitter.com/9FwCt020yG
— Databricks (@databricks) 26 janvier 2026[DEMO] Databricks AI Guardrails block sensitive data before it ever leaves your environment. @dennylee shows how PII like UK National Insurance numbers and credit card data is detected and stopped locally at the serving endpoint, instead of being sent to an external model.
-
AI Agents Pose Dangers: Safety Research and Dialogue Essential
By
–
AI agents can be dangerous. Safety and security must come first. Caution, research and dialogue are needed.
-
The challenge of model transparency and black box neural networks
By
–
Yeah, they open sourced it, but most of the work is done by a neural network and they didn't share the weights of that. So it's still 2/3rds a black box. And that black box is changing every month (I think it actually is changing faster than that).
-
Anthropic prepares Security Center for Claude Code with manual scans
By
–
Anthropic is preparing to release Security Center (formerly AutoPatch) for Claude Code. Users will be able to browse historical scans and detected issues, as well as trigger new scans manually. Will it be a direct response to the upcoming Codex upgrades and cybersecurity
-
Limitations of Intelligence: Laplace’s Gremlin and Irreducibility
By
–
New blog post on some of the intuitions I have on ways of describing limitations on intelligence. https://
inverseprobability.com/2026/01/25/lap
laces-gremlin-and-irreducibility
… It also links to a new arXiv paper that continues to look at the inaccessible game … a sandbox for idea formalisation. -
Cognitive Cost of Convenience: AI and Brain Health
By
–
The Cognitive Cost of Convenience: What Happens to Your Brain When You Outsource Thinking to AI. #AI #Learning #Brain #Brainhealth #GenAI #TechNews #Technology