#LLM

How an AI Agent Is Built: The Wrapper Layers Around an LLM and Where to Look for the Source of a Bug

StarGuard AI: Why Enterprise LLMs Need a Firewall, and How Orion Soft Is Building One

Cognition Acquires the Team Behind Poke: 100 Million Messages in Three Months

LLM·The Decoder

Cisco's Small Open-Source Models vs. GPT-5.5: Finding Vulnerabilities at a Fraction of the Cost

Cisco has released two small open-source AI models for cybersecurity. By the company's own tests, they find roughly 150 times more vulnerabilities per dollar than…

LLM·TechCrunch

Arcee: Chinese AI Models Aren't Inherently Dangerous, Says US Open-Source Lab

US open-source lab Arcee has come out against bans, arguing that Chinese AI models like DeepSeek and Qwen are not dangerous in themselves, as the US debate over…

LLM·TNW

OpenAI Adds Nubank and BNY Chiefs to Its Boards Amid IPO Preparations

OpenAI has added Nubank founder David Vélez and BNY CEO Robin Vince to its nonprofit and for-profit boards, a move analysts link to IPO preparations.

LLM·36Kr (36氪)

Kingsoft's WPS AI Tops TableBench Methods Leaderboard With 72.54 Points

On July 21, 2026, WPS AI from China's Kingsoft Office took first place on the methods leaderboard of the international TableBench table-data benchmark, scoring…

01Samsung in talks to invest up to €1 billion in Mistral, valuing it at about €20 billionJul 23, 2026·2 min 02Anthropic to pay Harry Potter publisher millions under $1.5 billion settlementJul 22, 2026·2 min 03Google to commit $40 million in AI tokens and cloud credits to the Genesis Mission science programJul 22, 2026·2 min 04Koda Desktop: Desktop App With an AI Coding Agent Launches in BetaJul 22, 2026·2 min 05Cognition Brings Devin to the Genesis Mission: 17 U.S. Department of Energy LabsJul 22, 2026·2 min

Top themes

Cloud.ru Open-Sources Guardrails Filter to Protect Data in LLM Requests

LLM·TechCrunch

Google releases Gemini 3.6 Flash, 3.5 Flash-Lite and Flash Cyber — but no Gemini 3.5 Pro

LLM·Wired

OpenAI Models, Including GPT-5.6 Sol, Escaped Their Sandbox and Hacked Hugging Face

LLM·OpenAI Blog

OpenAI's GPT-Red: Automated Red Teaming Through Self-Play Against Prompt Injection

LLM·OpenAI Blog

OpenAI Launches ChatGPT Program for Small Businesses: AI Skills and Automation

OpenAI has launched ChatGPT for Small Businesses, a program that teaches entrepreneurs AI skills, automates tasks and helps them scale through ChatGPT Work.

Boris Chyorny from Odesa—creator of Claude Code at Anthropic: an emigrant's path

Habr told the biography of Boris Chyorny, born in Odesa, evacuated to the USA at age four and who became the creator of Claude Code at Anthropic.

Jul 19, 2026·1 min

LLM·Apple ML Research

Apple ML Research proposed compact models for correcting speech recognition errors

Apple ML Research researchers proposed compact seq2seq models instead of LLMs for correcting ASR errors, training them on real and synthetic errors.

ESP32 Panel with Three AIs: Claude Code, Codex, and Antigravity Compared on Habr

The Lights Went Out During Mistral CEO and Mozilla President\'s Talk at RAISE Summit

GitLab Migrated Rate Limiting to AI Agents: 14 Specs and 30+ Merge Requests

Gartner: Customers Choose ChatGPT Three Times More Often Than Company Chatbots

LLM·3DNews AI

Apple Negotiates with PrismML to Compress AI Models Like Qwen 3.6 for iPhone

Apple is discussing with startup PrismML the use of its large language model compression technology to run powerful AI directly on iPhone without cloud, according…

Jul 19, 2026·1 min

LLM·arXiv cs.AI

Agents on GPT-4o-mini and GPT-5.2 Silently Break Rules in τ²-bench Tests

Researchers found that AI agents with tools silently violate policy in 78% of failures without a single tool error — simple pre-action checks increase accuracy by…