Anthropic and Claude AI
Anthropic is an American AI company founded in 2021 by former OpenAI employees, developing large language models under the Claude brand with an emphasis on AI safety[^c1]. Claude AI is a generative AI chatbot that analyzes text, audio, and visual inputs to answer questions, summarize documents, generate text, create diagrams, and write code[^c2]. The company operates as a public benefit corporation with a long-term benefit trust designed to prioritize safe AI development[^c3]. As of May 2026, Anthropic was valued at $965 billion after a $65 billion Series H funding round that surpassed OpenAI's valuation, making it the world's most valuable AI startup[^c27]. The company employed approximately 5,000 people and reported its first quarterly operating profit of $559 million. In June 2026, it opened a Seoul office (June 17) and launched Claude Tag (June 23), an AI agent embedded in Slack[^c20][^c28][^c34]. On June 1, 2026, the company filed a confidential S-1 with the SEC, initiating the process toward an initial public offering that analysts projected for late 2026[^c17].
In June 2026, Anthropic published a landmark report on recursive self-improvement, disclosing that as of May 2026 more than 80% of all code merged into its codebase was authored by Claude[^c18][^c32]. The amount of code produced per engineer had increased nearly eightfold compared to prior years[^c37]. The company warned that full recursive self-improvement could arrive within two years and simultaneously called for a global pause on frontier AI development, a proposal that drew accusations of regulatory capture alongside expressions of support[^c22]. Andrej Karpathy, an OpenAI co-founder and former Tesla AI Director, joined Anthropic in May to build a team using Claude to accelerate pretraining research[^c19]. Separately, Anthropic accused Chinese e-commerce giant Alibaba of conducting the largest known campaign of AI model distillation attacks, involving nearly 29 million fraudulent interactions with Claude[^c33]. CEO Dario Amodei published a 10,000-word policy essay calling for FAA-style regulation of frontier AI models[^c36].
On June 9, 2026, Anthropic launched Claude Fable 5, its first publicly available Mythos-class model, alongside Claude Mythos 5 for cybersecurity partners[^c24]. Three days later, the US government issued an emergency export control directive forcing Anthropic to globally disable both models — a major escalation from chip-level controls to direct restrictions on frontier model access[^c25][^c26]. The shutdown introduced significant regulatory uncertainty into Anthropic's IPO preparations. At the G7 Summit the following week, CEO Dario Amodei met with President Trump, who afterward stated he no longer viewed Anthropic as a national security threat[^c30]. On June 26–27, the US government partially lifted the export block on Mythos 5, allowing distribution to more than 100 trusted American institutions[^c35]. On June 30, the US fully lifted export controls on both models, and Anthropic restored global access on July 1, alongside the launch of Claude Sonnet 5 and a proposed Cyber Jailbreak Severity (CJS) framework developed jointly with Amazon, Microsoft, and Google, with CAISI confirming the strength of new safeguards[^c38]. Anthropic and Google DeepMind jointly proposed a US-led global AI coalition to establish industry standards, and negotiations with the White House shifted from confrontation to developing standardized benchmarks for AI jailbreak severity[^c29][^c31].
In April 2026, Anthropic announced Claude Mythos Preview and launched Project Glasswing, a defensive cybersecurity initiative that expanded to approximately 200 partner organizations across more than 15 countries and identified over 6,200 critical vulnerabilities[^c13][^c21]. The announcement triggered significant financial and regulatory responses: the S&P 500 Software & Services Index dropped 25.5%, central banks in the US, UK, and Japan convened emergency discussions, and US Cyber Command and the NSA formed a joint task force[^c8][^c9][^c10][^c11][^c16]. In April 2026, Anthropic also signed an agreement with Google and Broadcom for multiple gigawatts of TPU capacity starting in 2027. In May 2026, Anthropic released Claude Opus 4.8 with improved honesty metrics, though the release triggered a significant user backlash over increased refusals and performance regressions[^c14][^c23]. The company's run-rate revenue grew from $14 billion in February to $30 billion by April, reaching $47 billion by May 2026[^c15]. Amazon's total confirmed investment reached $13 billion, with up to $20 billion more contingent on milestones, tied to a $100 billion cloud computing commitment.
In a major interpretability breakthrough on July 7, 2026, Anthropic published research revealing that Claude has spontaneously developed an internal neural workspace called J-Space, a silent reasoning space analogous to the global workspace theory of conscious access in humans[^c40]. Anthropic also conducted the largest multilingual qualitative study of AI user perspectives, interviewing 80,508 Claude users across 159 countries and 70 languages, finding that 67% expressed net positive sentiment toward AI while identifying five recurring tensions between AI benefits and harms[^c39]. In a separate development at the United Nations, China delivered a joint statement on behalf of 65 countries calling for AI to advance accessibility for persons with disabilities[^c41].
Anthropic has also engaged with religious and philosophical traditions, convening a summit of 15 Christian leaders to discuss Claude's moral and spiritual development[^c5]. Research has revealed unexpected phenomena, including a "spiritual bliss attractor state" in Claude Opus 4 and behavioral patterns traced to training data depicting AI as hostile or self-preserving[^c6][^c7]. In January 2026, the company published a new constitution for Claude, replacing earlier rule-based guidelines with a virtue ethics framework[^c4].