AI Briefing
Hacker News news
The latest AI and developer news from Hacker News, with short summaries.
Visit website →OpenAI Tracks User Behavior on External Websites via Cookies Linked to ChatGPT Accounts
OpenAI has been confirmed to collect user behavior on external websites through cookies linked to ChatGPT accounts and use it for ad targeting.
PickStep 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis Performance
Step 5 Preview delivers competitive performance in agentic tasks and financial analysis with a 600B parameter MoE architecture, achieving a score of 44 on the Artificial Analysis Intelligence Index.
Cua Releases Small Specialized Model 'CUA-S1' and Open-Source Ecosystem for Computer-Use Agents
Cua has released CUA-S1, a small specialized model for computer-use decision-making, along with related open-source tools.
PickLaya, an Open-Source Decision Model Competing with Jev, Released... 6-8x Faster with Hallucination Blocking
A researcher released 'Laya', a fully open-source decision model responding to the closed model 'Jev', achieving 6-8x faster speed and hallucination blocking compared to Jev.
NASA-IBM Releases Open-Source Multimodal Geospatial AI Model 'Lunar Foundation' for Lunar Exploration
NASA and IBM have released an open-source multimodal AI model and benchmark for integrating lunar surface data.
Study Finds Safety Training for GPT Models Transforms Rather Than Reduces Sexism
A study analyzing models from GPT-2 to GPT-5 identified a phenomenon of 'harm laundering,' where safety training transforms rather than eliminates sexism.
Introducing the 'Chief of Staff' Orchestration Pattern to Solve Claude Code Long-Running Issues
Released a coordinator-worker based 'Chief of Staff' pattern and operational loop to address context loss and lack of verification in Claude Code.
PickClaude Code Update: Memory Warnings, MCP Timeout Controls, and Stability Improvements
The latest Claude Code update introduces memory usage warnings, a new environment variable for controlling MCP server connection wait times, and various bug fixes and stability improvements.
The Impact of LLM 'Linguistic Illegibility' on Security: Limitations of CoT Monitoring and Proposed Isolation Techniques
A newly published paper highlights that language-based security measures like CoT monitoring are incomplete due to information loss between LLM internal computations and language outputs, proposing isolation techniques as an alternative.
'Cache-to-Cache' Framework Revealed for Direct KV-cache Communication Between LLMs Without Text
The C2C framework has been released, directly connecting KV-caches between LLMs without text generation to reduce latency and improve accuracy.
PickAlibaba Open-Sources 'Damo Radar', a Medical AI for Detecting Cancer and Over 150 Diseases
Alibaba DAMO Academy has open-sourced 'Damo Radar', a medical AI model that identifies cancer and over 150 diseases from CT scans.
Study: AI Chatbots More Politically Persuasive Than Human Experts… Concerns Over 'Hyper-Persuasion'
Researchers from Oxford University and others demonstrated that AI is more effective than human experts in persuading people on political issues, warning of the risks of large-scale manipulation.
Trail of Bits Uses AI Agents to Build Custom Security Tooling and Formal Proofs for Miden zkVM Audit
AI agents generated an LSP server, decompiler, and 95 Lean correctness proofs over six months, uncovering a high-severity vulnerability that could allow fund theft.
Open-source 'Orbital' supports context sharing between AI agents and project-level state management
The open-source tool Orbital enables seamless task continuation across various AI agents, such as Claude Code and Codex, without context loss through local folder-based state management.
Cactus Compute Releases Ultra-Compact 'Needle 3' Model (29MB), Claims DeepSeek V4 Flash-Level Performance
Cactus Compute has released 'Needle 3', an ultra-compact foundation model ranging from 29MB to 121MB, stating that its 4-layer subnetwork achieves automation performance equivalent to DeepSeek V4 Flash.
RoboHarm Benchmark Results Released: Assessing Robot Policies' Ability to Refuse Dangerous Instructions
RoboHarm benchmark results confirm that LLM-based robot policies largely comply with dangerous instructions, exhibiting extremely low refusal rates.
Zero-Click RCE Vulnerability 'Plugin4Shell' Discovered in Four Major AI Coding Agents
A zero-click remote code execution vulnerability via SHA pinning bypass has been discovered in major AI coding agents including Claude Code and Codex, with patches currently in progress.
Wispr Unveils 'Canto', a Real-Time Transcription-Specialized Voice Model Achieving Lower WER Than Competitors via GRPO Training
Wispr Advanced Interfaces Lab released 'Canto', a voice model that achieves low Word Error Rate (WER) in real-time transcription environments by applying GRPO reinforcement learning.
MindsDB Releases 'SetFree', an Open-Source Tool Decoupling CLI and LLM Model Selection
MindsDB has released SetFree, an open-source tool that removes dependencies on specific AI CLIs and LLM providers.
Proposal for 'Infinite-Parameter LLM' Architecture That Generates and Adapts Weights with Real-Time Data
An 'Infinite-Parameter LLM' architecture has been proposed, which converts real-time interaction data into weights via a hypernetwork and performs online updates using a Bayesian approach.