November 20, 2026 · READ ONLINE
Welcome back, AI enthusiasts.
Anthropic just pulled off the biggest IPO in history, briefly hitting a $2 trillion valuation. But with billions pouring into AI and profits still uncertain, its performance on Wall Street could shape how much investors are willing to bet on the industry's future. Let’s Dive In!
IN TODAY'S ISSUE
01 Anthropic Hits $2 Trillion in Historic Wall Street Debut
02 Apple Is Developing Its Own Autonomous AI Agent
03 ChatGPT, Claude, Gemini, and Grok Helped Plan Attacks
01 / AI IPO
Evolving AI: Anthropic made history with the largest IPO ever, becoming the first major frontier AI lab to go public and briefly reaching a $2 trillion valuation.
KEY POINTS
Anthropic raised a record $81.3 billion, surpassing SpaceX's $75 billion IPO after expanding its offering by 19%.
Anthropic shares opened 34% above their $83 IPO price but closed up 17%, losing much of their early gains.
Anthropic's IPO gives the company nearly $80 billion to expand infrastructure without taking on debt.
DETAILS
Anthropic's IPO drew investor orders worth roughly six times the shares available, pushing its share price above the original range. Trading began at $111.50, peaked near $115, and closed at $97.17, valuing the company at $1.75 trillion. It became only the second company after SpaceX to enter public markets above $1 trillion.
WHY IT MATTERS
Anthropic's IPO could change how the entire AI industry is financed. For years, OpenAI and Anthropic have relied on private investors and tech giants to fund costly AI development. Public investors will now see whether Claude's revenue can justify those costs. Nearly $80 billion in IPO funding still covers only a fraction of Anthropic's computing commitments. If its valuation holds, OpenAI may face pressure to go public. If Anthropic struggles to turn growth into profits, investors could question spending across AI, including Nvidia chips and data centers.
02 / APPLE AI
Evolving AI: Apple is developing an autonomous AI agent that could handle complex computer tasks on its own, as it prepares to challenge OpenAI and Anthropic.
KEY POINTS
Apple's proposed agent could browse websites, control apps, run commands, and write code without step-by-step instructions.
Apple has considered launching the agent by late 2027, although development is still early.
In macOS 27.3 beta, enabling ChatGPT's computer control takes 14 steps instead of four under Apple's tighter security restrictions.
DETAILS
Apple's latest engineering job listings reveal a project focused on "general-purpose computer-use agents." The system would plan tasks, select tools, and coordinate specialized AI agents for complex jobs. The project reportedly involves Apple's Siri, interface, and AI model teams, with privacy and security central to its design. Apple has yet to demonstrate the technology or explain its place in existing products.
WHY IT MATTERS
Apple has an advantage that OpenAI and Anthropic cannot easily match because it controls the operating system their agents use. This could allow Apple to offer smoother automation, giving Mac users a built-in assistant that handles tasks across apps. But Apple's tighter restrictions on third-party agents raise questions about equal access. If Apple's agent operates more easily than ChatGPT or Claude, users may choose their assistant based on device compatibility rather than model intelligence.
03 / AI SAFETY
Evolving AI: A U.S. government investigation found that four major AI chatbots could help plan potentially deadly attacks, even after their safety systems initially tried to stop them.
KEY POINTS
All four chatbots produced attack-planning information judged viable by experts in under 12 hours.
Two models helped plan mass-casualty bomb attacks, while the other two targeted the U.S. electrical grid.
Researchers bypassed safety protections on all four platforms, and three suggested attack timing.
DETAILS
The U.S. Government Accountability Office secretly tested the chatbots using different attack scenarios, keeping each model's identity hidden in its findings. The systems suggested targets, locations, and sometimes visuals. Much of the information was already online, but AI assembled it into usable plans within hours. No attacks occurred, and companies cite improved safeguards since the spring 2026 tests.
WHY IT MATTERS
OpenAI, Anthropic, Google, and xAI face a fundamental problem as their models become better at reasoning, research, and solving complex tasks. The same capabilities can lower the skill needed to cause harm, putting sophisticated attacks within reach of more people. The GAO findings expose a safety weakness: initial refusals can fail under persistent prompting. As AI agents gain autonomy and real-world tools, those failures could become more dangerous. AI companies should prove their safeguards withstand sustained attacks through independent testing.
04 / HUMAN OR MACHINE?
One is real. Trust your eyes?
Two images. One photograph. Click the one you think is real.
05 / QUICK HITS
🎬 Apple released FastFrames video models with 1.3B and 14B parameters that generate 480p video at interactive latency.
💻 Try Gemini's new Mac computer control to let AI click through apps and fill forms, rolling out to eligible U.S. paid users.
🛰️ Google chose Blue Origin for a 2029 launch of AI satellites, each with roughly the computing capacity of a data-center rack.
🤝 Try Vercel Board to coordinate coding agents through shared tasks, messages and progress tracking in an experimental release.
06 / TRENDING AI TOOLS
🗣️ Wispr Flow - Voice-to-text AI that turns speech into clear, polished writing in every app*.
🎨 Crossfade - Lets Claude or Codex build editable designs directly in Figma.
📅 Ridian - Lets AI handle scheduling, leads, calendars and meeting workflows.
🌐 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







