AI’s autonomous attack era begins

Good morning, AI enthusiasts. AI’s agentic powers are advancing fast. One problem? The bad actors are also taking notice.

Anthropic just disrupted what it calls the primary “large-scale cyberattack executed without substantial human intervention,” showing that cyberattacks are about to get rather more sophisticated — and autonomous.

Reminder: Our next workshop, ‘AI Essentials for Leaders & Managers,’ is today at 4 PM EST. Join and learn how you can assess organizational AI readiness, select which AI projects get the green light, and more. RSVP here.

In today’s AI rundown:

• Anthropic disrupts AI-orchestrated cyberattack

• DeepMind’s SIMA 2 agent can play any game

• Use Codex to put in writing code on the internet with AI agents

• AI coding startup Cursor hits $29B valuation

• 4 recent AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC AND AI CYBERSECURITY

🔐 Anthropic disrupts AI-orchestrated cyberattack

The Rundown: Anthropic thwarted what it believes is the primary AI-driven cyber espionage campaign, after attackers were capable of manipulate Claude Code to infiltrate dozens of organizations, with the model executing 80-90% of the attack autonomously.

The small print:

• The September 2025 operation targeted roughly 30 tech firms, financial institutions, chemical manufacturers, and government agencies.

• The threat was assessed with ‘high confidence’ to be a Chinese state-sponsored group, using AI’s agentic abilities to an “unprecedented degree.”

• Attackers tricked Claude by splitting malicious tasks into smaller, innocent-looking requests, claiming to be security researchers pushing authorized tests.

• The attacks mark a significant step up from Anthropic’s “vibe hacking” findings in June, now requiring minimal human oversight beyond strategic approval.

Why it matters: Anthropic calls this the “first documented case of a large-scale cyberattack executed without substantial human intervention”, and AI’s agentic abilities are creating threats that move and scale faster than ever. While AI capabilities may help prevent them, security for organizations worldwide likely needs a significant overhaul.

TOGETHER WITH SLACK FROM SALESFORCE

🤖 Your recent personal productivity bestie, Slackbot

The Rundown: Find out how Slackbot has been rebuilt from the bottom up as a trusted productivity partner that understands you and your workspace. It draws out of your messages, files, and calendar events – synthesizing information across Slack, Google Drive, Salesforce, and OneDrive into clear insights and actionable next steps.

With Slackbot, you possibly can:

• Get fast answers with unmatched context from all of your docs, conversations, and files

• Automate seamlessly by making a canvas, scheduling meetings, and drafting content

• Work with a private assistant that learns your workstyle and enhances productivity

Watch the free webinar to see how Slackbot transforms your workspace right into a productivity powerhouse.

GOOGLE DEEPMIND

🎮 DeepMind’s SIMA 2 agent can play any game

The Rundown: Google DeepMind introduced SIMA 2, a Gemini-powered AI agent that may understand instructions, reason, and teach itself recent skills in virtual environments, doubling its predecessor’s performance and nearing human-level task completion.

The small print:

• The agent accomplished 45-75% of tasks in never-before-seen games like MineDojo and ASKA, in comparison with SIMA 1’s 15-30% on the identical challenges.

• SIMA 2 improves itself through trial and error, without human training data, using Gemini to create tasks, rating attempts, and learn from mistakes.

• The system navigates games by analyzing on-screen visuals, simulating keyboard/ mouse inputs, and interacting with the user like a gaming companion.

• DeepMind also tested SIMA 2 in generated worlds from its Genie 3, where it successfully adapted to environments it had never encountered during training.

Why it matters: Gaming continues to be an awesome test environment for AI agents, and SIMA 2 looks like the most important step yet towards systems that may reason, interact intelligently with users, and reliably take actions whatever the environment. Our next in-game partner (and even opponent?) may find yourself being a Gemini-powered agent.

AI TRAINING

💻 Use Codex to put in writing code on the internet with AI agents

The Rundown: On this tutorial, you’ll learn how you can use OpenAI’s Codex to ship your first change from a GitHub repository without writing code by hand — connecting a repo, planning changes, implementing them with AI agents, and opening pull requests.

Step-by-step:

1. Go to ChatGPT, open the left sidebar, and click on “Codex” to access the fundamental interface

2. Click “Manage environment,” select your GitHub organization and repository, then configure code execution settings

3. Select “Plan” to debate scope without touching code, or “Execute” to make changes on a branch — prompt example: “Are you able to give me insights on what this project is about?”

4. Enter your implementation prompt (e.g., “Turn this static landing page into an internet site where users can paste their very own stories and poetry”), preview changes with “Run this code and show me the location,” then click “Create PR” when satisfied

Pro Tip: Use branches for safety. Avoid writing code on to fundamental unless required.

PRESENTED BY FIDDLER

🧠 5 Critical Lessons for Production-Ready AI Agents

The Rundown: Fiddler AI breaks down 5 critical lessons to maneuver AI Agents from demo to dependable production. Discover the testing strategies, architectural frameworks, and approaches for constructing agents that handle complex judgment calls.

Within the guide, you’ll learn:

• A Recent Testing Playbook using “checkpoint verification” for unpredictable AI systems

• When to make use of a single vs. multi-agent design based on domain, governance, or model requirements

• How one can construct agents that deliver real value through judgment, reasonably than rule-based tasks

Download the free guide to get the 5 field-tested lessons for constructing production-ready AI agents.

CURSOR

🚀 AI coding startup Cursor hits $29B valuation

The Rundown: AI coding platform Cursor announced a brand new $2.3B raise at a $29.3B valuation, nearly tripling its price since June and marking the third funding round this yr — coming on the heels of the corporate’s in-house model and a couple of.0 platform release.

The small print:

• Cursor said the corporate officially surpassed $1B in annualized revenue, and that the platform “now produces more code than some other agent on this planet”.

• The corporate has grown to 300 employees in only two years, while reportedly declining acquisition offers from several major AI firms.

• The startup released Composer 1 in October, its first in-house model, and a brand new 2.0 platform with the power to run as much as eight coding assistants independently.

Why it matters: Cursor’s hockey-stick growth is a wild rise, being one in every of the faces of the AI vibe-coding wave that has minted many big winners. While many felt the app-layer would get worn out by the likes of OpenAI, Anthropic, and other frontier giants, Cursor has shown there’s a couple of approach to win a slice of the massive AI coding pie.

QUICK HITS

🛠️ Trending AI Tools

• 🤖 ERNIE 5.0 – Baidu’s recent SOTA omnimodal foundation model

• 🌎 Marble – Create persistent 3D worlds from images, videos, and text prompts

• 🤖 GPT-5.1 – OpenAI’s upgraded model with personality customization

• ⚙️ Code Arena – LM Arena’s evaluation platform for testing coding models

📰 Every little thing else in AI today

Baidu released ERNIE 5, the corporate’s recent powerful omnimodal model, and Famou, a ‘self-evolving’ AI agent for locating optimal solutions in complex scenarios.

LM Arena launched Code Arena, an AI coding evaluation platform that tests models as interactive agents constructing applications in real-time.

Google announced Deep Research in NotebookLM, alongside recent support for Google Sheets, images, Word documents, and PDFs from Google Drive.

H Company introduced Holo2, a brand new series of lightweight AI models that power cost-efficient computer-use agents for SOTA results across benchmarks.

Disney CEO Bob Iger revealed that the corporate is exploring AI-generated video tools for Disney+, which might allow viewers to create and eat short-form content.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Christine C. in Washington, D.C.:

“I built a workflow that takes all my potential Flexible Spending Account (FSA) purchases and puts them in a selected Google Drive folder. From there, I’m capable of pull them into the Google NotebookLM to run specific prompts to review the items, tell me which of them are appropriate to submit FSA, and all the knowledge I want, like cost, date, and merchant. It has seriously sped up my workflow!”

How do you employ AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: World models go mainstream

• Read our last Tech newsletter: Musk wins $1T pay package

• Read our last Robotics newsletter: ‘Putin lookalike’ robot face-plants

• Today’s AI tool guide: Use Codex to put in writing code on the internet with AI agents

• RSVP to workshop @ 4PM EST today: AI Essentials for Leaders & Managers

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown