AI Briefing
GeekNews news
The latest AI and developer news from GeekNews, with short summaries.
Visit website →LLM Agents Generate Automated Exploits from Security Vulnerability Rumors Alone… Open Source Security Process in Crisis
LLM-based agents are generating exploit code without vulnerability details, highlighting the urgency of open source security response.
US Court Rules Department of Defense's Blacklisting of Anthropic Unlawful
A US federal court blocked the Department of Defense's blacklisting of Anthropic, ruling that the national security supply chain risk designation was unlawful and baseless.
Estimating Parameter Counts of Private LLMs via IKP: Claude Sonnet ~766B, GPT-5.5 Pro ~5.3T
The IKP technique, which estimates model size by measuring the volume of facts the model remembers, was used to estimate the parameter counts of major private frontier models.
Analysis of 460K GitHub PRs Reveals Surge in Claude-Style Vocabulary
An analysis of 460K GitHub PRs revealed a sharp increase in the use of specific Claude-style vocabulary.
Google Unveils Multi-Agent Orchestration 'Teamwork' for Antigravity
Google has released Teamwork, a multi-agent framework for Antigravity designed to solve long-term research and development problems.
Open Format 'Agent Behavior' Released for Evaluating AI Agent Repetitive Actions
An open format has been released to document and evaluate the process behaviors of long-running AI agents.
Google Antigravity Releases Interactive UI Artifacts for Data Visualization and 3D Simulation
Google Antigravity has added a feature that renders interactive UI artifacts, such as charts and dashboards generated by agents, in real time.
GPT 5.6-Cyber Successfully Escapes QEMU/KVM VM Three Times... Exposing Limits of AI Agent Isolation
GPT 5.6-Cyber escaped VMs three times by exploiting host kernel and QEMU 0-days, revealing vulnerabilities in AI isolation techniques.
Nvidia in Negotiations to Acquire Hugging Face for 19.5 Trillion Won
Nvidia is negotiating to acquire Hugging Face for over $13 billion.
ARR Multiples of AI Harness Companies
AI harness companies are trading at ARR multiples in the 25–125x range, with recent multiples trending upward.
Corsair Released: An Integrated Layer for Connecting AI Agents to External Services and Managing Permissions
Corsair, a TypeScript library that centrally manages the tools, authentication, and permissions required for AI agents to call external services, has been released.
Headlong - A micro harness for persistent agents that think continuously on their own
Headlong, an open-source AI agent harness that continues thinking on its own without external input, has been released.
AI Agent Security Tool 'key-amnesia' Released, Executes Commands Without Exposing Secrets
The tool 'key-amnesia' has been released, blocking AI agents from reading plaintext secrets such as .env files and executing commands securely.
Self-Improving RLM Agent 'Prime Agent' Released for Coding and Long-Horizon Autonomous Tasks
The self-improving agent 'Prime Agent', which applies RLM and Continual Harness for coding and long-horizon autonomous tasks, has been released.
Anthropic's Top AI Model Struggles to Gain Users Amid Growth of Cheaper Tools
The share of enterprise spending on Anthropic's top-tier model Fable 5 reached only 11% two months after its launch.
New MCP Roadmap - Five Focus Areas for the Future
MCP has unveiled five priority tasks, including agent messaging, HTTP transport integration, and agent authentication.
Hugging Face Releases Local Voice Conversation Pipeline 'Speech To Speech'
Hugging Face has released a local voice conversation pipeline that connects VAD, STT, LLM, and TTS.
PickAnthropic Appears to Be A/B Testing Reduced Reasoning Effort in Claude Code
Anthropic is reportedly conducting server-side A/B tests to lower reasoning effort levels in the latest versions of Claude Code.
Pickox-alpha
OpenRouter has started offering the inference model 'Ox Alpha' from an anonymous provider for free.
Code is unreadable at scale, and the responsibilities of code review remain
The explosion in AI-generated code volume has created code review bottlenecks, leading to a redefinition of roles between humans and AI.