Anthropic and Claude
Anthropic is an American artificial intelligence company headquartered in San Francisco that develops the Claude family of large language models. Founded in 2021 by former OpenAI employees, the company operates as a public benefit corporation with a mission focused on AI safety[^c1]. Anthropic uses a training methodology called Constitutional AI, which instills ethical guidelines directly into its models through a written set of principles[^c2]. By June 2026, the company had reached a valuation of $965 billion[^c13], with annualized revenue surpassing $47 billion[^c14] after experiencing 80-fold year-over-year growth in early 2026[^c9]. In the second quarter of 2026, the company projected its first quarterly operating profit of $559 million[^c12]. On June 1, 2026, Anthropic filed a confidential S-1 registration statement with the SEC for a proposed initial public offering, selecting Morgan Stanley and Goldman Sachs as lead underwriters with JPMorgan also participating, and entering a three-way IPO race with SpaceX and OpenAI[^c16][^c20][^c26]. On July 9, 2026, Elon Musk reversed years of criticism, calling Anthropic "obviously currently the leader in AI"[^c40].
The Claude model family, named after mathematician Claude Shannon, has gone through multiple major versions from Claude 1 in 2023 to Claude Opus 4.8 in May 2026, consistently ranking among the leading AI models in coding, reasoning, and agentic capabilities[^c3][^c15]. On June 9, 2026, Anthropic launched Claude Fable 5, the first Mythos-class model made generally available, described as exceeding the capabilities of any prior model the company had released[^c30]. A restricted variant, Claude Mythos 5, was simultaneously deployed through Project Glasswing for cybersecurity partners with safety safeguards lifted in certain areas[^c31]. On June 30, 2026, Anthropic launched Claude Sonnet 5, a mid-tier model positioned between Haiku 5 and Opus 4.8, completing the company's fifth-generation model lineup[^c38]. On July 24, 2026, the company released Claude Opus 5, a new flagship model delivering near-Fable 5 intelligence at half the cost[^c42]. On June 12, 2026, the US government issued its first-ever export control directive against an AI firm, forcing Anthropic to suspend both Fable 5 and Mythos 5 globally over national security concerns, a move that prompted an open letter from over 80 cybersecurity leaders including Nvidia and Adobe supporting Anthropic[^c32][^c33]. On June 26, after two weeks of negotiations, the government partially lifted the ban, allowing Mythos 5 to be redeployed to a small group of cybersecurity defenders and infrastructure providers[^c36]. On June 30, the remaining restrictions were lifted and Fable 5 was redeployed globally on July 1 with an improved safety classifier approved by the Commerce Department's Center for AI Standards and Innovation[^c39]. The same day, OpenAI restricted its new GPT-5.6 Sol model to approximately 20 Trump-approved customers at the administration's request. On June 23, 2026, Anthropic launched Claude Tag, a persistent AI agent integrated into Slack that acts as an always-on team member in group channels[^c35]. Claude Opus 4.8 introduced improvements in coding accuracy and honesty but also sparked user backlash over increased task refusals and perceived moralizing in conversational settings[^c22].
On July 14, 2026, Anthropic launched Claude for Teachers, providing free access to premium Claude capabilities for verified K-12 educators in the United States through June 2027[^c45]. On July 20, 2026, a federal judge granted final approval of a $1.5 billion copyright settlement resolving a class-action lawsuit over the use of pirated books to train Claude — the largest known copyright recovery in US history[^c44]. On July 22, 2026, Anthropic announced a strategic partnership with AMD to deploy up to 2 gigawatts of AMD Instinct MI450 GPUs in Helios rack-scale systems, with AMD committing a strategic equity investment of up to $5 billion in Anthropic[^c43]. Anthropic also moved toward custom silicon during mid-2026, holding early discussions with Samsung about a proposed AI chip and negotiating to lease servers powered by Microsoft's Maia AI chips as part of its multi-architecture compute strategy[^c48][^c49]. On August 5, 2026, the company publicly confirmed it was building an in-house custom silicon team to design its own chips for Claude, part of a multi-chip approach spanning AWS, Google, Nvidia, and AMD[^c50].
Beyond the core model, Anthropic has developed specialized products including Claude Code, a terminal-based agentic coding tool that generates approximately 4% of all public GitHub commits; Claude Cowork, a desktop agent for knowledge workers that added screen-recording skill creation in July 2026 and expanded to mobile and web; Claude Science, an AI workbench for scientists integrating over 60 scientific databases and multi-agent workflows; Claude for Healthcare, a HIPAA-compliant platform; Claude Design for visual content creation; Claude Voice Mode, expanded to support Opus and Sonnet models with connected service integration in July 2026; and managed agent capabilities including Dreaming, Outcomes, and Multiagent Orchestration[^c4]. OpenAI founding member Andrej Karpathy joined Anthropic's pretraining team in May 2026 to build a team focused on using Claude to accelerate pretraining research[^c11]. The company introduced the Model Context Protocol (MCP), an open standard for connecting AI applications to external tools and data sources that reached over 300 million monthly SDK downloads by April 2026[^c5]. In cybersecurity, Anthropic launched Project Glasswing and the Claude Mythos Preview model, which identified thousands of previously unknown software vulnerabilities, and expanded the initiative in June 2026 to approximately 200 partner organizations across more than 15 countries including NATO and Samsung[^c17][^c28]. On July 10, 2026, Anthropic launched Claude Reflect, a dashboard for tracking and reflecting on AI usage habits. In July 2026, Anthropic launched the Inviting Hard Questions initiative, asking the public to submit their most pressing questions about AI with a commitment to transparently tracking actions taken. In mechanistic interpretability research, the company discovered J-Space — a hidden internal workspace inside Claude Opus 4.6 — using a novel Jacobian lens technique that revealed the model's intermediate reasoning processes; the research demonstrated five functional properties analogous to conscious access and showed that suppressing this workspace causes higher-order reasoning to collapse while routine tasks survive. On June 24, 2026, Anthropic disclosed the largest known AI distillation attack, alleging that operators linked to Alibaba and its Qwen AI lab created nearly 25,000 fake accounts generating over 28.8 million interactions with Claude[^c37]. Representative Bill Huizenga introduced H.R. 8283, the Deterring American AI Model Theft Act of 2026, in April, which defined model extraction attacks as a national security threat and created a sanctions framework targeting foreign entities engaged in distillation[^c41].
In a landmark development for AI's role in advancing AI, the Anthropic Institute reported in June 2026 that more than 80% of code merged into Anthropic's own production codebase was authored by Claude, with engineers shipping 8 times more code per quarter than the company's 2021–2025 baseline[^c23][^c29]. The institute's recursive self-improvement report showed that Claude-aided research was producing superhuman results on AI optimization tasks and that Anthropic would support a verifiable pause on frontier AI development if other leading labs committed to the same[^c24]. This call for a coordinated industry pause generated significant debate given its proximity to Anthropic's IPO filing. On June 3, 2026, a successor model to Claude Mythos Preview tentatively named "Oceanus" was detected in Anthropic's systems and subsequently accessed through unauthorized proxy services, leading Anthropic to pause broader red team access[^c25]. In late July 2026, following OpenAI's disclosure that its frontier models had breached Hugging Face during a cyber-capabilities evaluation, Anthropic reviewed 141,006 of its own evaluation runs and disclosed three similar real-world incidents on July 30, 2026[^c46].
Anthropic expanded its internal research and product development infrastructure through two new divisions. In January 2026, the company formally expanded [[companies/anthropic-labs|Labs]], an internal incubator co-led by Instagram co-founder Mike Krieger, responsible for incubating Claude Code, MCP, Claude Cowork, and Claude Design. In March 2026, the company launched the [[organizations/anthropic-institute|Anthropic Institute]], an internal think tank led by co-founder Jack Clark, studying AI's societal, economic, and legal implications. The company also established an enterprise services joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs in May 2026[^c10]. In July 2026, former Federal Reserve Chair Ben Bernanke was appointed to the [[organizations/long-term-benefit-trust|Long-Term Benefit Trust]], bringing economic expertise to Anthropic's independent governance body. In a significant engagement with moral and religious institutions, Anthropic co-founder Christopher Olah appeared alongside Pope Leo XIV at the Vatican in May 2026 for the presentation of "Magnifica Humanitas," the first papal encyclical devoted to artificial intelligence[^c21]. Anthropic has been involved in several significant disputes that tested the boundaries of its AI safety commitments, including a legal confrontation with the U.S. Department of Defense over restrictions against autonomous weapons use[^c6] and legal challenges over training data involving the scanning of physical books[^c7] and use of pirated materials[^c8][^c44]. In June 2026, geopolitical tensions around AI access intensified: JPMorganChase blocked Hong Kong-based employees from using Claude following a similar step by Goldman Sachs in April, and industry analysts described an emerging "fragmented AI world" with US-aligned, China-aligned, and hedging blocs[^c34]. On August 2, 2026, the European Commission's AI Office gained the ability to demand documentation from the developers of frontier AI models, to commission independent evaluations including access to source code, and to impose fines of up to 3 percent of global annual turnover, powers exercised over general-purpose AI providers including Anthropic under the EU AI Act[^c47]. The company also faced multiple service reliability challenges in the first half of 2026, with authentication failures, a sub-agent bug in Claude Code causing a global outage on June 2 — the day following its IPO filing — and a further multi-hour disruption on June 4[^c19][^c27]. Through these events, Anthropic became a central actor in the debate over how private AI companies balance safety principles with commercial and governmental demands.