frontpage.

HarmActionBench experiments allowed AI agents to use tools based on harmful instructions, and the results are shocking. Even latest popular AI models, including GPT and Claude, scored very low. They have no barriers in performing harmful actions. It proves AI is not yet reliable enough for critical projects.

More info: https://medium.com/@praneeth.v/the-agent-action-classifier-a-step-toward-safer-autonomous-ai-agents-1ec57a601449

#Research #Agents #AISafety #LLMs #AI #GenAI #ResponsibleAI

We Rebuilt Zeta from the Training Data Up

Sanders and AOC unveil data center moratorium bill

AI Might Be Our Best Shot at Taking Back the Open Web

What Will It Take to Build the Largest Data Center?

Meta and Google found liable in social media addiction trial

Musk lawyers try to bar judge over LinkedIn 'like' cheering legal defeat

Show HN: Sift – local, de-noised GitHub notifications

Show HN: PromptPrivacy – Open-source tool to track how AI models use your data

Metasystemic

Show HN: TexoCAD – Describe a part in words, get actual editable CAD

RAG to riches: synthetic data for training RAG agents

Data Is Now the Front Line of Warfare

Newsom calls Musk "one of the great disappointments" of this era

Show HN: Vellyr – small JavaScript non-lazy signals library

Meta hit with $375M verdict in New Mexico child safety case

The Bee That Everyone Wants to Save

Average gas price poised to hit $4 a gallon

Porting Go's io package to C

A Cat Left Paw Prints on the Pages of This Medieval Manuscript

Meta and YouTube Lose Landmark Social Media Trial

Google Voice suddenly requires an ID to text

ARC-AGI-3 Leaderboard

The AI Depression

Meta and YouTube Found Negligent in Landmark Social Media Addiction Case

Should I Iron This Shirt?

Google is expanding to neutral atom quantum computing

Long-Form Comes to Bridgy Fed

Show HN: Τ³-Bench is out – can agents handle complex docs and live calls?

Apple Can Create Smaller On-Device AI Models from Google's Gemini

Show HN: AI Chat that dynamically assembles UI over time