Anthropic CEO Dario Amodei Getty Images for HubSpot Claude maker Anthropic confidentially filed paperwork with the Securities and Exchange Commission, taking the first step towards its highly anticipated initial public offering that could come as early as fall 2026. The AI behemoth just raised an eye-popping $65 billion funding round that valued it at $965 billion, topping OpenAI’s valuation and minting Anthropic as the most valuable AI startup in the world. Anthropic announced that its annualized revenue crossed $47 billion in May, up from $30 billion earlier this year, thanks to growth from its marquee AI coding tool Claude Code and a powerful and only partially released AI cybersecurity model called Mythos Preview. In April, it launched Project Glasswing, an initiative that gave some 50 companies like Apple, Google and JP Morgan Chase access to Mythos Preview to find and patch security vulnerabilities in their systems. (The companies uncovered some 10,000 critical security flaws, Anthropic said.) The company announced on Tuesday that it is expanding access to the model to over 150 organizations across 15 countries to safeguard their software. As businesses ramp up spending on AI tools like Claude, they’re facing so-called “AI sticker shock”—ballooning costs with little productivity gain to show for it. One company reportedly spent $500 million in a single month after failing to put limits on their employees’ Claude usage, Axios reported. Microsoft canceled Claude Code licenses for its employees, encouraging them to rely on homegrown applications instead. While strong enterprise adoption has been the basis for Anthropic’s revenue growth, companies cutting back on AI spending could slow its roll. Even rival OpenAI has highlighted rising costs as a major concern, with CEO Sam Altman calling wasteful AI usage one of the most “fair criticisms” of AI right now. Let’s get into the headlines. BIG PLAYS OpenAI is transforming its coding tool Codex into a general-purpose application for knowledge workers. The AI juggernaut announced today that it’s adding a swath of abilities to the app for different types of work including data analysis, product design, sales, creative production and investment banking. The company also announced that more than 5 million people now use Codex every week, with non-coders making up a fifth of its user base. HUMANS OF AI Venture capitalist Sarah Guo bet everything on AI before ChatGPT mainstreamed the technology, leaving a cushy job at Greylock to launch AI-focused VC fund Conviction in 2022. The gamble has clearly paid off. Guo, who made her debut on the Forbes Midas list this year, has backed a number of high-flying AI startups: legal tools builder Harvey; ChatGPT for doctors OpenEvidence and AI coding agent startup Cognition. “There are a lot of priors that come from traditional venture about both markets and how you build companies that we thought would be challenged,” Guo tells Forbes. AI DEALS OF THE WEEK Base Power, an Austin-based startup developing home batteries, is in talks to raise around $1 billion at a $12 billion valuation, sources familiar with the deal told Forbes. Cofounded by billionaire Michael Dell’s son Zach Dell in 2023, the company aims to provide backup power to homes during outages, addressing grid reliability problems thanks to the rapid expansion of AI data centers. Also notable: AI security startup Gray Swan, which pressure tests models for every major frontier lab including OpenAI and Anthropic, raised $40 million in Series A funding at a $200 million valuation, Forbes reported. DEEP DIVE Matthew Harvey Sanders still remembers how confusing it was to become Catholic. Raised Protestant, he became curious about Catholicism thanks to a high school crush and in college decided to convert. The process was grueling, filled with constant questions: Why do Catholics care so much about Mary? Why speak to saints instead of directly to God? Google often provided conflicting answers, and Sanders couldn't exactly call his priest every time a new question arose. When ChatGPT launched in November 2022, Sanders saw a business opportunity in a conversational chatbot focused on helping people understand Catholicism. The following year, Sanders launched Magisterium AI, an LLM-based bot that can answer complex theological questions. The system relies on existing Gemini and GPT models, but is trained on over 32,000 Roman Catholic doctrines and teachings spanning 2,000 years. “There's no way any priest would be able to have the information we have in our database inside of their heads,” says Sanders, 44. “Even the best theologians in the world, it’s not fair to expect them to act like a machine.” Within three weeks of launch, Sanders noticed a surprising shift in user behavior. Though he’d originally built the product as a research tool for people interested in converting, students, scholars and priests, users began asking highly nuanced moral and personal questions. That pushed him to start developing his own Catholic AI model, Ephrem, expected to launch in 2027. The goal is for Ephrem to identify Church teachings and saints relevant to a user’s question, then suggest specific readings, prayers, habits or virtues to cultivate—what Sanders describes as putting a saint in your pocket. The Catholic Church is increasingly taking a public role in the AI debate. This week, Pope Leo XIV presented his first major social encyclical—an official letter on moral and social issues—warning that AI is already reshaping human relationships, institutions and power. He criticized the use of AI in warfare, political manipulation and image distortion, and warned that the technology could become a new form of exploitation. That concern is not limited to the Vatican. Anthropic, now valued at $965 billion, has been actively seeking religious input on how Claude should respond to moral and spiritual questions. In March, the company invited 15 Christian leaders to its San Francisco office to discuss how chatbots should handle personal questions, including those involving relationships, faith and self-harm. Anthropic cofounder Chris Olah, who is described as an atheist, spoke in Rome after the Pope presented the encyclical, saying AI is powerful, but poorly understood and in need of moral guidance. Read the full story on Forbes. MODEL BEHAVIOR Over the weekend, hackers took over popular Instagram accounts like The Obama White House and Sephora by tricking Meta’s AI-powered customer support assistant into granting access to targeted profiles. The hackers used a VPN to match their location with that of their victim, then asked Meta’s AI assistant to update the account’s email address. From there they could reset the password and hijack the account. Meta said the issue has now been resolved, but the hack underscores how AI features spawn entirely new security vulnerabilities that hackers can exploit.

Source: Forbes

Read Original Source →