November 17, 2026 · READ ONLINE
Welcome back, AI enthusiasts.
China’s AI labs are using answers from OpenAI and Claude to train cheaper rivals. That could mean lower AI prices and more choices for users, while putting pressure on the American companies spending billions to build these models. And if OpenAI and Anthropic tighten access to protect their technology, developers and businesses could face new restrictions on the AI tools they rely on. Let’s Dive In!
IN TODAY'S ISSUE
01 China Is Using American AI to Build Its Own
02 Trump Steps Into Jensen Huang's Nvidia Comeback Fight
03 Google Uncovers AI Agents Built to Hack on Their Own
01 / CHINESE AI TRAINING
Evolving AI: An investigation reveals Chinese developers using OpenAI and other American models to train cheaper rivals as U.S. chip restrictions tighten.
KEY POINTS
Over 300 Chinese documents show the practice spreading from universities into commercial AI development.
Researchers turned 5,000 answers from OpenAI's o1 into 10,750 training examples to improve Alibaba's Qwen2.5.
U.S. officials warned Chinese regulators in July that large-scale collection of American AI outputs could amount to intellectual-property theft.
DETAILS
DeepSeek and Moonshot used GPT-6 Astra to train newer models, according to people with direct knowledge. Anthropic also accused Alibaba of extracting Claude's reasoning to improve Qwen. Distillation teaches smaller models skills using answers from stronger systems, potentially cutting costs. Beijing encourages it across industry, though not specifically with U.S. models. Training still needs powerful chips, and some uses may violate providers' rules.
WHY IT MATTERS
OpenAI and Anthropic risk helping rivals catch up using technology they spent heavily to develop. Chinese labs can reuse AI-generated answers, coding solutions and reasoning examples to teach rival models similar skills without accessing the original code or weights. As distillation improves, U.S. leaders could lose pricing power while still funding costly frontier research. Restricting access might slow the practice but hurt legitimate customers. American companies could end up funding breakthroughs while cheaper competitors win their customers.
You deserve a CRM that matches your team’s ambition.
Attio is the CRM for agentic revenue. Build it around your business and run any revenue play, from founder-led sales to your first team of AEs.
The Attio Startup Program gets you 80% off the annual Pro plan for your first year and 30% off your second. You’ll get dedicated onboarding support, plus perks from Linear, Notion, Intercom and more.
Modal, Granola, Railway and Wispr Flow are already building on Attio.
02 / JENSEN COMEBACK
Evolving AI: President Trump joined Elon Musk in backing Jensen Huang's return as Nvidia CEO, while Sam Altman praised Huang but stopped short of endorsing his comeback.
KEY POINTS
Trump argued Huang learned a lesson and questioned unproven Chinese espionage allegations.
Musk backed a petition demanding Huang's return, now with over 180,000 signatures.
Altman called Huang a hero but lacked enough information to judge whether he should return.
DETAILS
President Trump's backing intensified the fight over Huang's return after his sudden resignation amid allegations about personal relationships and potential national security concerns involving businesswoman Alice Huang. Despite pressure, Nvidia's board reportedly isn't considering reinstatement, while Treasury Secretary Scott Bessent favors an outsider and Colette Kress remains interim CEO. Altman's caution carries personal weight after his brief removal as OpenAI CEO in 2023 before employees and investors rallied to bring him back.
WHY IT MATTERS
Trump and Musk turned Nvidia's crisis into a test of who controls one of AI's most important companies. Reinstating Huang before the investigation concludes could weaken the board's authority, while refusing means proving Nvidia can protect the partnerships behind its dominance in AI chips. Altman's caution highlights the stakes for companies like OpenAI, whose growth depends on Nvidia staying stable through the transition.
Y Combinator explores how AI coworkers are taking on real tasks, from building software through Slack to managing ad budgets. In this hour-long conversation, Grok Bot's builders explain how to assign agents responsibilities, control their access, and catch mistakes before they cause problems.
03 / HACKER AI
Evolving AI: Google uncovered a hacking system designed to let AI agents find security weaknesses, develop exploits, and launch cyberattacks with little human involvement.
KEY POINTS
Google documented its first case of hackers using AI frameworks designed to fully automate finding and exploiting software vulnerabilities.
Google found stolen victim data and attack scripts, with evidence pointing to French media, entertainment, and education websites.
AI hacking frameworks have already been used in real intrusions, including against U.S. aerospace organizations.
DETAILS
Google found exploit_agent on an exposed server, connected to several advanced AI models. The system compares software code before and after security updates to identify weaknesses, then automatically writes and runs attacks against vulnerable versions. The server also contained tools for exploring breached networks and installing hidden access, although the number of successful autonomous attacks remains unknown.
WHY IT MATTERS
Google's discovery could change the economics of cybercrime, allowing small hacking groups to launch sophisticated attacks that once required entire teams. While AI agents can hunt for weaknesses around the clock, businesses still need time to test and install security fixes. That growing speed gap could make even short delays costly, pushing companies toward AI defenses that can detect and stop attacks before customer data is compromised.
04 / HUMAN OR MACHINE?
One is real. Trust your eyes?
Two images. One photograph. Click the one you think is real.
05 / QUICK HITS
🎨 Google upgraded Pics to Nano Banana 2.1 for more accurate image edits and clearer text in Docs and Slides.
🧠 Try Hugging Face's new model routing to access supported open AI models across 80+ providers with one token.
🤖 Download n8n's Nod agent for Mac or Windows to automate tasks across websites and apps using your preferred AI model.
📱 Join Apple's health AI study if you are 21+ in the U.S. with an eligible iPhone, with enrollment is capped at 200,000.
06 / TRENDING AI TOOLS
🎞️ Guideless - AI tool that automatically turns your clicks and workflows into step-by-step video guides*.
🔖 Marqly - Lets you ask questions across everything you have bookmarked.
📝 Yogurt - Free open-source AI meeting notes that stay private and local.
🌐 Spacefast - Lets your existing AI build, publish and update web apps.
*PARTNER LINKS
BEFORE YOU GO
Worth your four minutes?
Help us make the next edition better.
See you tomorrow,
The Evolving AI team









