JadidJust-in AI Developments, Incidents & Deadlines
October 2026
SMTWTFS12345678910111213141516171819202122232425262728293031

AI news, Thursday, October 8, 2026

14 stories

TLDR

  • Tensorlake npm SDK hijacked to spread the Shai-Hulud credential worm; it plants config files that re-run when a repo is opened in Claude Code or VS Code (Oct 7–8) Security Agents
  • Unpatched CVSSCommon Vulnerability Scoring SystemA standard 0 to 10 severity score: 0.1–3.9 low, 4.0–6.9 medium, 7.0–8.9 high, 9.0–10 critical.FIRST ↗ 9.8 RCERemote code executionA flaw that lets an attacker run malicious code on a target's computers or network.Cloudflare ↗ in LMCache (the LLMLarge language modelA neural network trained on huge amounts of text that generates text, answers questions and summarizes.Anthropic ↗ KV-cache layer used with vLLM); no fixed version (Oct 7) Security Infra
  • Microsoft ships RTX Spark AI PCs and a revamped Windows 11 with "Execution Containers" to sandbox AI agents (Oct 7) Agents Infra
  • Manus raises $500M+ after Meta's acquisition was unwound (Oct 8) Agents
  • OpenAI launches "Intelligent UI" in ChatGPT: interactive charts, calculators and diagrams in answers (Oct 7) Customer AI

Action Items

Things your engineering team should look at or do this week.

  • Check for Tensorlake exposure. Search lockfiles and CI logs for tensorlake@0.5.144. If found, rotate npm, GitHub, AWS and SSH credentials, then look for unexpected .claude/settings.json, .vscode/tasks.json and GitHub Actions workflows in your repos.
  • Find any LMCache deployments. The multiprocess server must not be bound to a routable address (JFrog's advice). Keep it on localhost or a trusted cluster network until a fix ships.
  • Audit internet-facing LLM infrastructure (LiteLLM proxies, Gotenberg, Gitea). A crypto-mining botnet is exploiting known bugs in exactly these. Patch and remove public exposure.
  • Review agent filesystem permissions on Macs. Apple plans to require explicit user action for Full Disk Access. Check which of your agent tools request it and whether they really need it.

Models & Tools

  • Microsoft Surface Laptop Ultra, Surface RTX Spark Dev Box and revamped Windows 11 (Oct 7) Agents Infra — Built on Nvidia's RTX Spark chip. The laptop starts at $2,600 (top model sold out), the dev box at $6,000 and ships with VS Code, GitHub Copilot CLI and WSL. Windows 11 adds "Execution Containers" for sandboxing AI agents, which Nadella says will reach all Windows 11 users. Dell's XPS 16 Creator Edition is $3,800 for late-October delivery. (TechCrunch)
  • Nous Research launches "Hermes for Businesses" (Oct 7) Agents Open Source — Deployable custom agents for multi-step workflows with private data, built on its open-source Hermes agent. It raised a $90M Series B at a $1.5B valuation. The "2.5% of global tokenA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗ usage" and "24M clones" figures are company estimates. (TechCrunch)
  • OpenAI "Intelligent UI" for ChatGPT (Oct 7) Customer AI — Answers can include tappable buttons, custom calculators, interactive charts and editable graphs. Pro, Plus, Business and Enterprise users got it Oct 7; Free and Go users on Oct 8. (TechCrunch)
  • Google AI Edge "Foresight" Mac meeting-notes app (Oct 8) Customer AI Data — A Granola competitor that can run fully offline using on-device models (Gemma 4 assistant, EmbeddingGemma 2 embeddings) and builds a knowledge base from your files. Mac-only, optimized for Apple Silicon; pricing not stated. (TechCrunch)
  • Google Labs "Playground" prompt-to-game platform (Oct 7) Customer AI — Browser games from text prompts, 2D or 3D, shareable by link. US-only, 18+, weekly token allowance. (TechCrunch)

Security

  • Tensorlake npm package compromised (Shai-Hulud / ChainDrop campaign) (Oct 7–8) Agents — Malicious tensorlake@0.5.144 (since removed) steals npm, GitHub, AWS, Kubernetes and Vault credentials and self-propagates by republishing the victim's packages. It also writes config files so it re-runs when a project opens in Claude Code or VS Code, and has a destructive "hostage token" fallback. If installed, remove it and rotate every credential. (The Hacker News)
  • LMCache unauthenticated RCE, CVE-2026-105192 (CVSS 9.8) (Oct 7) Infra — Affects 0.3.9 through 0.5.5 (latest stable) plus 0.5.6 release candidates; no fixed version exists. JFrog says to keep the multiprocess server off routable addresses; a firewall reduces but does not remove the risk. (The Hacker News)
  • PoeLLM botnet infects 3,400+ servers via exposed AI/LLM infrastructure (Oct 7) Infra — The "Canto Incognito" campaign exploits known flaws in public LiteLLM, Gotenberg, Gitea and Ivanti Sentry deployments to install miners. It hides its C2 address in a poem on GitHub. No remediation guidance in the article. (The Hacker News)
  • Apple to tighten macOS Full Disk Access over AI agent risk (Oct 5) Agents Policy — FDA will require explicit user action; no ship date. It follows a report that Meta's Muse agent read a journalist's iMessages, and a since-patched Muse Mac flaw that exposed its auth token. (The Hacker News)
  • AgenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ zero-dayAn attack using a flaw that was unknown before it was exploited.NIST ↗ chain breaches DIVD's Zammad ticketing system (Sep 21, reported Oct 1) — The Dutch vulnerability-disclosure institute says the evidence points to an AI agent driving the attack (CVE-2026-102489 and CVE-2026-102490 chained to root); the agent made noisy mistakes along the way. Older than a week but newly circulating in this week's ThreatsDay recap. (The Hacker News)

Industry & Business

  • Manus raises $500M+, first round since Meta's $2B deal was unwound (Oct 8) Agents — Led by Boyu Capital and IDG Capital, with Tencent and HSG participating. Valuation undisclosed (reported at $4B in talks in September); a Hong Kong IPO is reportedly under consideration. (TechCrunch)
  • Tab, iMessage/WhatsApp personal assistant, exits stealth at $300M valuation (Oct 7) Agents Customer AI — Texts-based assistant with its own phone number, computer and wallet; irreversible actions need user approval. Funding details undisclosed. (TechCrunch)
  • Vesta raises $30M for agent swarms in mortgage lending (Oct 8) Agents — (TechCrunch)

Policy & Safety

  • Google launches synthid.com to check media for AI watermarks (Oct 7) Policy — Accepts images, video and audio; detects only SynthID-watermarked content (Google tools, plus partners including OpenAI and NVIDIA). Google says it sees about 1M verification requests a day. Microsoft and Meta use separate, non-interoperable schemes. (TechCrunch)

Type to search · ↑↓ to move · enter to open