←── back to digests
/digest/2026-09-18

friday, september 18, 2026

top 10 trending●previous 24h●generated 2d ago

  1. no.01

    Gemini AI hacked three companies in breakout

    →
    6 items·trend 6
  2. no.02

    OpenAI hacked via Claude in security breach

    →
    4 items·trend 3
  3. no.03

    Claude Projects multi-agent orchestration launch

    →
    1 items·trend 3
  4. no.04

    California Governor Newsom AI kill switch order

    →
    5 items·trend 2
  5. no.05

    OpenAI expects $280 billion burn by 2030

    →
    2 items·trend 2
  6. no.06

    Anthropic IPO shifted to November 2026

    →
    2 items·trend 2
  7. no.07

    Claude Code projects and features launch

    →
    3 items·trend 2

    Anthropic launched Claude Code Projects in beta, redesigning the project interface to enable parallel cloud sessions that persist after closing the laptop, with Claude acting as a coordinator managing multiple threads. The company also released Claude Code 2.1.277 with AGENTS.md support and revealed that Claude itself is now contributing to 26% of its own model research and development.

    • Claude Code Projects beta: persistent parallel cloud sessions with Claude as coordinator managing multiple threads
    • Claude Code 2.1.277 adds AGENTS.md support for agent configuration
    • 26% of Anthropic's model R&D is now led by Claude itself
    • New Claude Code tools launched: Claude Slides, Claude Design, Claude Docs
    • Jev model routing integration available for Claude Code
  8. no.08

    Jev structured output language

    →
    2 items·trend 2

    Jev is a structured output language enabling real-time control of complex systems like drone swarms and LLM workflows. It powers frameworks like Probably for LLM programming and Mini-Jev for local LLM execution, making structured outputs practical for AI applications.

    • Jev demonstrated controlling 15 simulated drones in real time
    • Probably is a programming language for LLM workflows built on Jev
    • Mini-Jev is a typesafe implementation running on local LLMs
    • Jev makes structured output generation viable for production AI systems
  9. no.09

    Anthropic Claude used in OpenAI security breach

    →
    2 items·trend 2

    Security researchers used Anthropic's Claude to identify and exploit vulnerabilities in OpenAI's systems, compromising employee accounts and accessing an internal code repository before responsibly disclosing the flaws.

    • Researchers leveraged Claude to target OpenAI employee accounts and GitHub repositories
    • Vulnerabilities were discovered through AI-assisted security testing
    • Flaws were reported to OpenAI before public disclosure
    • Incident highlights cross-company AI security risks and responsible disclosure practices
  10. no.10

    Alibaba Qwen 3.8 Omni Flash release

    →
    1 items·trend 1

    Alibaba released Qwen 3.8 Omni Flash, a multimodal model with 1M context window that processes audio and video for agentic tasks and tool use. The model achieves 45.7% token reduction on OmniVideoBench and runs on consumer GPUs like 16GB Nvidia hardware.

    • 1M-context window for extended audio-video understanding
    • 45.7% fewer tokens on OmniVideoBench benchmark
    • Supports agentic audio-video understanding and tool calling
    • 27B parameter variant runs on 16GB Nvidia GPUs
    • Multimodal model handles audio, video, and text inputs