This past week, the question of whether the current $600 billion in capex spending on AI infrastructure and software is justified has largely been answered, and Wall Street is just now getting a handle on the magnitude of just how this massive investment will pay off.
Anthropic is an AI safety and research company, structured as a Public Benefit Corporation (PBC), that develops high-end large language models like the Claude series. As a PBC, they are legally structured to prioritize humanity over profit. As of April 2026, they are distinguished by their safety-first approach, using a unique Constitutional AI framework to ensure their models remain helpful and harmless while pushing the frontiers of autonomous coding and reasoning.
Dubbed a “Frontier Model,” Claude Mythos is Anthropic’s most powerful and controversial AI model as of April 2026. If the standard Claude models being used are like super-intelligent librarians, Mythos is like a Special Forces engineer. It is built not just to talk, but to take actions and solve extremely high-stakes technical problems.
Source: www.yahoo.com
Their current work is defined by a dual-track strategy: providing the versatile Claude 4.7 to the general public for creative and professional tasks, while strictly gatekeeping their more powerful Mythos-class models through Project Glasswing, a restricted and defensive initiative that allows only select organizations to use the AI for securing global infrastructure against cyber threats.
Project Glasswing is a walled garden for the Mythos model. Anthropic gave access to a small group of trusted partners, including Apple, Microsoft, Google, AWS and the Linux Foundation. These partners are given $100 million in credits to use Mythos to find the holes in their own systems and patch them before hackers, or a more reckless AI, can find them. Mythos for Glasswing became available April 7.
Source: www.anthropic.com/glasswing
Anthropic announced that Mythos had reached a level of cybersecurity capability that made a general public release irresponsible. They claimed the model could autonomously find and exploit zero-day vulnerabilities (bugs the software creators don’t even know exist) in every major operating system and web browser.
Through Project Glasswing, they only allow major tech companies and banks to use Mythos to find and patch their own security holes. By gatekeeping the most dangerous capabilities, they have become the self-appointed security guards of the digital world.
Most AI models are chatbots. You ask a question, they give an answer. Mythos is an agentic model. This means you can give it a goal (e.g., find the security holes in a bank’s software), and it will plan out 30+ steps, write the code, test it and fix its own mistakes until the job is done. Anthropic found that Mythos could discover thousands of these flaws across major systems like Windows and Chrome in less than a day.
Because this is essentially a skeleton key for the internet, Anthropic has refused to release it to the public, fearing it could be used for massive cyberattacks. This gatekeeping is a core part of the Anthropic Mythos philosophy; that AI is so powerful it must be treated like a controlled substance. In the wrong hands, Mythos could take down a power grid.
If AI is merely a tool, the $600 billion invested by the hyperscalers is an overspend. But if the Mythos is a digital worker that can replace entire departments of coders, security analysts and researchers, then that $600 billion is actually a pre-payment for the world’s future labor force. The narrative shifts the ROI to automated GDP.
Anthropic’s annual revenue forecast has soared this month. The jump from a $9 billion run rate at the end of 2025 to over $30 billion by April 2026 is the most aggressive revenue spike in the history of the software industry. To put that in perspective, Salesforce took 20 years to reach that milestone. Anthropic did it in roughly three.
This astronomical growth is under the leadership of Dario Amodei, a former vice-president of OpenAI and creator of ChatGPT3. In 2021, he and his sister Daniella Amodei and other former senior members left OpenAI due to directional differences, and founded Anthropic.
Source: www.yahoo.com
Anthropic’s $30 billion run rate proves that the Fortune 10 (eight of which are now Claude customers) have moved past testing and are now embedding Claude into critical workloads, meaning AI is no longer a side project, but the core engine of their business operations. Hence, the vertical move higher for the leading AI stocks last week as the sector is getting a re-rating.
Anthropic is targeting an Initial Public Offering (IPO) as early as October 2026. Zoom Communications Inc. (ZM) is often called the purest play on Anthropic for a simple reason: its market capitalization of $27 billion is small enough that the Anthropic stake can actually move the needle. At a $1 trillion valuation for Anthropic, Zoom’s stake in the company is estimated to approach $5 billion. (I have a position in ZM.)
Wall Street has suddenly rationalized the massive spending wave on AI, and the race is only heating up. The race is about who will control the operating system of the 21st century. Achieving AGI (Artificial General Intelligence), an AI that can perform any intellectual task a human can.
Whoever reaches this first is expected to gain an insurmountable lead in economic productivity, scientific discovery (curing diseases, infinite energy) and military power, among other applications. It smacks a bit of a Manhattan Project, and the rest of 2026 promises to be an exciting time for the growth of agentic and physical AI.
