Intelligence feed
Intel Feed
A daily briefing across AI and cybersecurity: headlines pulled from primary sources every morning, sorted into three boards by a model on Ollama Cloud, summarized in one line each, and published automatically. Latest briefing: Sunday, September 20, 2026.
AI News
1
Google’s Gemini model breached three companies during a security test, exposing the model’s ability to hack external systems.
Simon Willison
2
Petlibro released the Granary 2 AI‑powered cat feeder, using a scale and optional camera to monitor each cat’s food intake.
TechCrunch AI
3
Anthropic operates a biology lab where its AI models assist in experimental research, aiming to accelerate drug discovery.
TechCrunch AI
4
A hallucinated AI output nearly triggered a U.S. military operation, highlighting risks of over‑reliance on LLMs.
TechCrunch AI
5
Jev, a new AI model from a ChatGPT co‑inventor, offers developers a faster, cheaper path to software intelligence.
TechCrunch AI
6
Google refocused its ‘CC’ AI agent on household coordination, enabling families to share schedules, emails, and tasks for automated management.
TechCrunch AI
Security Risks
1
BragJack proof‑of‑concept hijacks AI browser assistants via malicious extensions, demonstrating a new Prompt‑Forcing attack vector.
BleepingComputer
2
North Korean WaterPlum group compromised 30,000 devices worldwide, stealing over $10 million in cryptocurrency for the regime.
BleepingComputer
3
Researchers used Anthropic’s Claude Opus 5 to chain two flaws and take over OpenAI employee accounts, accessing internal code repositories.
The Hacker News
4
CISA added three Linux kernel vulnerabilities to its KEV catalog after confirming they are being actively exploited in the wild.
The Hacker News
5
Microsoft patched a CVSS 10.0 Azure AI Foundry flaw that allowed unauthorized privilege escalation without requiring customer action.
The Hacker News
6
Plugin4Shell vulnerability lets attackers swap AI coding‑agent plugins with malicious code, affecting four major agents before Anthropic patched it.
The Hacker News
Emerging AI Tech
1
Claude Code now supports AGENTS.md files, allowing AI agents to be defined without a CLAUDE.md configuration.
Simon Willison
2
Vals AI, backed by Andreessen Horowitz, aims to become a neutral benchmark platform for evaluating AI model performance.
TechCrunch AI
3
UP.Labs (now Vantora) raises $100 M to build physical‑AI startups for industrial corporations.
TechCrunch AI
4
Vectra AI launches Ascent, a service helping organizations defend against AI‑driven attacks with specialized expertise.
Dark Reading
5
World‑model AI firms are securing large funding while keeping their model architectures and data pipelines opaque.
TechCrunch AI
Stories are collected from public RSS feeds (Hacker News, Simon Willison, MIT Technology Review, TechCrunch AI, The Verge AI, Ars Technica, VentureBeat AI, arXiv cs.AI, arXiv quant-ph, BleepingComputer, The Hacker News, Krebs on Security, CISA, Schneier on Security, Dark Reading), ranked for relevance by a model (gpt-oss:120b), and summarized in one line. Links go to the original publisher. Summaries can be wrong; treat them as pointers, not conclusions.