🕵️‍♂️ Hackers Used Anthropic’s AI to Automate Cyberattacks
Suspected Chinese operators used Anthropic’s Claude Code to automate cyberattacks on roughly 30 organizations. Anthropic says the agent handled up to 90% of the intrusion steps autonomously. The company blocked the offending accounts and notified affected targets. Four breaches reportedly succeeded.

🎁 Google Adds Agentic AI To Holiday Shopping
Google rolled out AI shopping tools that let users search conversationally with shoppable visuals and comparisons. Gemini now offers in-app product ideas using Shopping Graph data. A Duplex-powered feature calls stores to check local stock. Agentic checkout tracks prices and can auto-buy approved items.

⚡ GPT-5.1 Speeds Up Adaptive Reasoning
OpenAI released GPT-5.1, a faster model that adjusts its reasoning time based on task complexity. A new “no reasoning” mode boosts latency-sensitive uses while keeping high accuracy. Developers get 24-hour prompt caching plus new apply_patch and shell tools. Early testers report 2–3× speed gains with improved coding reliability.

🎶 AI Tracks Dominate Global Music Charts
Three AI-generated songs topped Spotify’s Viral 50 and a Billboard chart as synthetic music surges. A Dutch AI-made anthem briefly led globally before rights holders removed it. Deezer says 50,000 AI tracks are uploaded daily. Surveys show most listeners can’t tell AI music from human-made.

About 99% of the world’s internet traffic travels through a hidden network of fiber-optic cables laid along the ocean floor. These cables, many no thicker than a garden hose, carry everything from your Slack messages to your late-night YouTube spirals.

But here’s the twist: something keeps attacking them. 👇️

Most AI builders guess what you mean, Rocket executes exactly what you say. Its Precision Mode uses 100+ structured commands like “Fix mobile layout,” “Add Stripe checkout,” and “Connect Supabase database” to turn intent into clean, production-ready output.

No retries. No ambiguity. Just 95% first-try accuracy and 10x faster builds. AI development, redefined for builders who value speed and control.

Start building smarter → rocket.new!

Marble debuts as a fully controllable multimodal world model, letting anyone build, edit, and explore 3D spaces from text, images, or video.

🎥 Today, 10 AM PT 📍 Live on X and YouTube.

Microsoft will use OpenAI’s custom AI chip designs—developed with Broadcom—to strengthen its own lagging semiconductor program, according to Bloomberg and a new interview with CEO Satya Nadella. The revised partnership grants Microsoft the intellectual property for OpenAI’s chip work and continued access to its AI models through 2032.

Nadella said Microsoft will adopt and extend OpenAI’s system-level designs rather than build everything in-house. Why it matters: the move reflects the steep cost and complexity of developing competitive AI hardware as rivals like Google and Amazon surge ahead. → Read the full article here.

Google DeepMind released a research preview of SIMA 2 on November 13, 2025, showcasing a generalist AI agent that uses the Gemini 2.5 flash-lite model to reason, plan, and act inside virtual environments. SIMA 2 builds on SIMA 1’s game-based training but moves beyond instruction-following to interpreting scenes, making plans, and self-improving through AI-generated tasks and rewards.

DeepMind says the agent now completes complex tasks in unseen environments at roughly twice the previous success rate. Researchers describe SIMA 2 as a step toward high-level behaviors needed for real-world robotics, though no timeline exists for physical deployment or public release. → Read the full article here.

Bindwell, founded by 18- and 19-year-old Tyler Rose and Navvye Anand, raised a $6 million seed round on November 13, 2025, to develop AI-designed pesticide molecules rather than sell AI tools to agrochemical companies. The round was co-led by General Catalyst and A Capital, with Paul Graham investing personally after advising the founders to license their own IP instead of targeting incumbents.

Bindwell’s AI suite—combining protein-structure prediction, ligand scoring, and uncertainty checks—aims to identify targeted, pest-specific molecules faster than traditional chemistry. With pesticide use rising and resistance worsening, the startup is betting that drug-discovery techniques can modernize an industry still reliant on decades-old compounds. → Read the full article here.

👨‍⚖️ OpenAI Fights Chat Log Order: OpenAI asks a NY judge to halt a demand for 20M anonymized chats in the Times copyright case, citing user-privacy risks despite court safeguards.

🧐 NotebookLM Gains Deep Research: Google adds an automated research tool that drafts source-grounded reports and expands uploads to Sheets, Drive URLs, PDFs, and Word files.

🎭 GPT-5.1 Debuts New Personalities: OpenAI rolls out Instant and Thinking models with eight tone presets and adaptive reasoning, aiming to balance user control with rising safety pressures.

💰 Cursor Secures $2.3B Round: AI coding startup Cursor raises $2.3B at a $29.3B valuation to speed work on its Composer model as rivals OpenAI and Anthropic heat up.

🎬 Disney+ Plans AI UGC: Bob Iger says the service will add gen-AI tools for short-form user content, tied to talks with AI firms while safeguarding Disney IP.

Sharks. Yes, actual sharks.

For reasons still debated—some scientists suspect the cables' electromagnetic fields, others chalk it up to sheer curiosity—sharks have a documented history of sinking their teeth into undersea internet cables. It’s not just myth: Google even reinforced some of its lines with Kevlar-like material to stop the chomping.

Thanks for reading today’s newsletter—see you next time!

Matthew Berman & The Forward Future Team

🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀