Timeline of Anthropic

This is a timeline of Anthropic, an AI safety and research company focused on developing reliable and ethical artificial intelligence systems. Founded in 2021, it is known for creating Claude, a family of AI assistants designed for safety, interpretability, and alignment with human values.

Sample questions

The following are some interesting questions that can be answered by reading this timeline:

Big picture

Time period	Development summary	More details
2016-2020	Prelude	Before Anthropic's official founding, many of its future team members already conduct pioneering work in AI safety and alignment research at organizations like OpenAI. During this period, Dario Amodei (who would become Anthropic's CEO) leads safety teams at OpenAI, publishing influential papers on concrete problems in AI safety and alignment. This era sees the development of key techniques like reinforcement learning from human feedback (RLHF) that would later become foundational to Anthropic's approach. By late 2020, differences in research priorities and perspectives on AI development lead several researchers to plan a new organization more singularly focused on AI safety research and responsible AI system development.
2021-2022	Foundation period	Anthropic is founded in January 2021 by a team led by Dario Amodei and Daniela Amodei, along with several other researchers who previously worked at OpenAI. The company is established with a core mission of conducting AI safety research and developing reliable, interpretable, and steerable AI systems. During this initial period, Anthropic raises significant early funding (over $700 million), including investments from Dustin Moskovitz, Eric Schmidt, and others, allowing the company to build its research team and infrastructure.
2022-2023	Claude development and public introduction	This period sees the development and initial release of Claude, Anthropic's AI assistant. The first version of Claude is made available to select partners and researchers in late 2022, with broader beta access in early 2023. During this time, Anthropic focuses on developing its Constitutional AI approach, which uses reinforcement learning from human feedback (RLHF) and a set of principles ("constitution") to guide AI behavior. The company publishes important research papers on AI alignment and safety while continuing to refine Claude's capabilities.
2023-2025	Growth and product Expansion	Characterized by major product launches and significant growth, this period includes the public release of Claude 2 in July 2023, followed by the Claude API becoming widely available. In March 2024, Anthropic launches the Claude 3 model family (Haiku, Sonnet, and Opus), representing a significant advancement in capabilities. The company secures major partnerships and additional funding rounds, including investments from Google, Amazon, and others. Anthropic expands its product offerings beyond the chatbot interface to include the Claude API for developers and begin focusing on enterprise applications.

Full timeline

Year	Month and date	Event type	Details
2019		Departure from OpenAI	The Amodei siblings leave OpenAI due to directional differences regarding OpenAI's ventures with Microsoft.
2021		Company founding	Anthropic is founded by former OpenAI executives including Dario and Daniela Amodei. The company is classed as a Delaware public-benefit corporation, aiming to create reliable, interpretable, and steerable AI systems.
2021	May 28	Funding	During its first funding round, Anthropic raises $124 million in a Series A funding round to develop reliable and steerable AI systems. The funding supports computationally intensive research to advance general AI capabilities. The round is led by technology investor and Skype co-founder Jaan Tallinn, with participation from James McClave, Dustin Moskovitz, the Center for Emerging Risk Research, and Eric Schmidt. By this time, Anthropic actively hires researchers, engineers, and operational experts to further its mission.
2021		Founding	Anthropic is founded by seven former employees of OpenAI, including siblings Daniela Amodei and Dario Amodei.
2022	April 29	Funding announcement	Anthropic raises $580 million in a Series B funding round to advance steerable, interpretable, and robust AI systems. By this time, since its 2021 founding, the company had made progress in interpretability by reverse engineering small language models, improving steerability with reinforcement learning, and analyzing AI performance shifts. The funding supports infrastructure for safer AI models and policy research. CEO Dario Amodei emphasizes scaling AI safety, while President Daniela Amodei highlights governance and culture. The round is led by Sam Bankman-Fried and includes investors like Caroline Ellison and Jaan Tallinn.
2022	Summer	Development milestone	Anthropic finishes training the first version of Claude but does not release it, citing the need for further internal safety testing.
2022	December	Funding	Anthropic receives a total of $700 million in funding by the end of 2022, including $500 million from Alameda Research.
2023	February 3	Funding	Google invests $300 million in Anthropic, securing a 10% stake and initiating a strategic partnership. The collaboration involves Anthropic using Google Cloud’s advanced GPU and TPU infrastructure to scale and train its AI systems, including its assistant, Claude. Founded by ex-OpenAI researchers, Anthropic emphasizes AI safety and interpretability through methods like Constitutional AI. CEO Dario Amodei states that Google’s support would enable broader deployment of Anthropic’s models and strengthen their shared goal of building trustworthy AI. This alliance positions both firms to compete more effectively in the rapidly advancing generative AI landscape.
2023	March 8		Anthropic outlines their views on AI safety and rapid AI progress. They argue that AI's impact may rival the industrial and scientific revolutions and could emerge within a decade. They emphasize concerns about AI alignment, safety risks, and potential disruptions to society. Anthropic believes AI progress is driven by increasing computation, scaling laws, and improved algorithms, leading to systems surpassing human capabilities. They advocate for a multi-faceted, empirical approach to AI safety, urging collaboration among researchers, policymakers, and institutions to mitigate risks while ensuring AI’s responsible and beneficial development.
2023	March	Model Release	Anthropic introduces Claude, a next-generation AI assistant designed to be helpful, honest, and harmless. After testing with partners like Notion, Quora, and DuckDuckGo, Claude becomes broadly available through a chat interface and API. It excels at tasks such as summarization, search, writing, Q&A, and coding. Two versions are offered: Claude, a high-performance model, and Claude Instant, a faster, cost-effective option. Partners highlight Claude’s capabilities: Quora integrates it into Poe for conversational AI, Juni Learning uses it for online tutoring, and Notion enhances productivity with Claude-powered AI. DuckDuckGo employs it for search answers, while Robin AI leverages it for legal document analysis. AssemblyAI utilizes Claude to improve audio transcription and conversation intelligence. Anthropic continues refining Claude for more reliable, user-friendly AI applications. Quora’s Poe app would see success with Claude, highlighting its strengths in creative writing and image understanding, with millions of daily user interactions.
2023	March 30		Anthropic launches the Claude App for Slack, offering AI-powered assistance for workplace collaboration. Claude can summarize Slack threads, answer questions, and generate structured data, making it a “virtual teammate” for various tasks. Users can interact with Claude in group channels by mentioning @Claude or via direct messages. Built with AI safety techniques like Constitutional AI, Claude enhances productivity while maintaining reliability. Slack’s SVP of Product praises its conversational abilities and context retention. While Claude has limitations, such as occasional errors and lack of internet access, Anthropic is committed to improving and responsibly deploying AI technology.
2023	April 18	Model Update	First major upgrade to Claude AI (version 1.3) is made public.
2023	April 20		Anthropic announces its support for increasing federal funding for the National Institute of Standards and Technology (NIST) to enhance AI measurement and standards. Effective AI regulation requires accurate assessment tools, and by this time NIST has a long history of developing such frameworks. Despite AI’s rapid advancement, NIST’s funding had stagnated, limiting its ability to evaluate risks and ensure safety. Investing in NIST would enable rigorous testing, increase public trust, support regulation, and foster innovation. Anthropic recommends a $15 million funding increase for FY 2024 to strengthen AI governance and complement broader regulatory efforts. Congress is urged to prioritize NIST’s role in AI oversight.
2023	April 26	Partnership	Anthropic announces a partnership with Scale to bring its conversational AI assistant, Claude, to enterprises. Scale customers can now build applications using Claude, leveraging Scale’s expertise in prompt engineering, model validation, and enterprise-grade security on AWS. This collaboration enables businesses to integrate proprietary data sources like Google Drive and Outlook while ensuring AI performance and reliability. Dario Amodei highlights the partnership’s role in responsibly scaling Generative AI. By combining Claude with Scale’s robust deployment tools, enterprises gain an AI-ready solution for real-world applications. Anthropic looks forward to expanding responsible AI adoption through this collaboration.
2023	May 9		Anthropic introduces Constitutional AI as a method to instill explicit values into language models via a predefined constitution, rather than relying on large-scale human feedback. This approach enhances transparency, scalability, and oversight while minimizing human exposure to harmful content. Claude, Anthropic’s AI assistant, follows principles derived from sources like the UN Declaration of Human Rights and AI safety research. The model is trained to be helpful, honest, and harmless, while avoiding toxic outputs and judgmental behavior. Constitutional AI allows iterative refinement of AI behavior, fostering ethical AI development and responsible deployment in real-world applications.
2023	May 11		Anthropic expands Claude’s context window from 9K to 100K tokens, enabling it to process around 75,000 words—equivalent to hundreds of pages of text—in under a minute. This allows businesses to analyze extensive documents, synthesize information, and answer complex questions more effectively than traditional search methods. Claude can review financial reports, legal documents, and even entire codebases. A demonstration shows it correctly identifying a single modified line in The Great Gatsby within 22 seconds. This enhancement improves Claude’s utility for summarization, risk assessment, and technical documentation, with API access becoming available for businesses and developers.
2023	May 16	Partnership	Anthropic partners with Zoom to integrate its AI assistant, Claude, into Zoom’s customer-facing AI products, enhancing reliability, productivity, and safety. The first integration is in the Zoom Contact Center, improving user experience and agent performance. Dario Amodei highlights the collaboration’s goal of bringing robust, steerable AI to workplaces. Zoom’s federated AI approach is expected to incorporate multiple models, including Claude. Additionally, Zoom Ventures invests in Anthropic, reflecting a shared vision for customer-centric AI solutions built on trust and security. This partnership aims to develop AI applications that effectively meet business and user needs.
2023	May 23	Funding	Anthropic raises $450 million in Series C funding led by Spark Capital, with participation from Google, Salesforce Ventures, Zoom Ventures, and others. The funding is aimed at supporting development of safe, reliable AI systems, including Claude. Anthropic aims to expand product offerings, advance AI safety research, and improve Claude’s capabilities, such as handling 100K context windows.
2023	June 13		Anthropic shares its response to the NTIA’s Request for Comment on AI Accountability, outlining key policy recommendations. They stress the urgent need for robust evaluation standards for advanced AI systems and propose increased government funding for model evaluations, the establishment of risk thresholds, and mandatory disclosures. Other proposals include pre-registration of large training runs, external red teaming before AI release, and advancing interpretability research. Anthropic also calls for clearer antitrust guidance to enable safer industry collaboration. These measures aim to foster a comprehensive AI accountability framework that supports safe, transparent, and beneficial AI development.
2023	July 11	Model release	Anthropic releases Claude AI version 2.0, which offers improved performance with longer responses, enhanced coding, math, and reasoning abilities. It now handles up to 100K tokens per prompt, enabling it to process large documents efficiently. Claude 2's safety features are strengthened, reducing harmful outputs by 2x compared to its predecessor. It excels in long-form content creation and reasoning, making it valuable for businesses like Jasper and Sourcegraph.
2023	July 25		Anthropic outlines its approach to securing frontier AI models, emphasizing that these systems must be protected with measures beyond typical commercial standards due to their strategic importance. The company recommends applying “two-party control” and secure software development practices like NIST’s SSDF and SLSA to all stages of AI development. Anthropic advocates for treating frontier AI as critical infrastructure, encouraging public-private cooperation, and potentially using procurement policies to enforce security standards. These efforts aim to prevent misuse, ensure model provenance, and promote responsible AI development while maintaining productivity and alignment with human values.
2023	July 26		Anthropic’s post details its work on “frontier threats red teaming,” a method to assess national security risks posed by advanced AI models. Focusing initially on biological threats, the team collaborated with biosecurity experts to evaluate how models might aid harmful activities, such as designing biological weapons. They found that while risks are currently limited, they are growing with model capability. The research led to effective mitigations, including safer training methods and output filters. Anthropic emphasizes the urgency of scaling this work, encouraging collaboration with governments and third parties to develop safeguards, monitor emerging threats, and build a robust safety framework.
2023	August 9		Anthropic releases Claude Instant 1.2, a faster and more affordable version of its AI model, which is available via API. This update brings notable improvements over Claude Instant 1.1, especially in math, coding, reasoning, and safety. Claude Instant 1.2 achieves higher scores on benchmarks like Codex (58.7% vs. 52.8%) and GSM8K (86.7% vs. 80.9%), while also providing longer, more structured, and better-formatted responses. It performs better in multilingual tasks, quote extraction, and question answering. Safety is also enhanced, with reduced hallucination and increased resistance to jailbreaks. Businesses can access Claude Instant 1.2 for a range of practical AI tasks.
2023	August 15		Anthropic announces a strategic partnership and $100 million investment from SK Telecom (SKT), Korea’s largest mobile operator. The collaboration aims to develop a customized large language model (LLM) tailored for telecommunications, using fine-tuning techniques to enhance Claude’s performance in telco-specific tasks like customer service, sales, and marketing. The multilingual model will support languages including Korean, English, Japanese, and Spanish. SKT experts will provide feedback to further refine Claude for industry applications. This partnership combines Anthropic’s AI capabilities with SKT’s telecom expertise, aiming to drive global AI innovation and leadership in the telco space through safer, more reliable AI solutions.
2023	August 23		Claude 2 becomes available on Amazon Bedrock. Bedrock is a managed AWS service offering access to top foundation models via API, simplifying generative AI integration. Claude 2’s availability enhances AI solutions across industries: LexisNexis uses it for legal AI services, leveraging its long-context processing; Lonely Planet employs it to unlock travel content for personalized planning; and Ricoh USA integrates it to generate training datasets while ensuring security and compliance. This collaboration aims to make safe, reliable AI more accessible for businesses globally.
2023	September 7	Product launch	Anthropic launches Claude Pro, a paid subscription for its Claude.ai chat experience, available in the US and UK. Priced at $20/month (US) or £18/month (UK), Claude Pro offers 5x more usage of the Claude 2 model compared to the free tier, enabling extended conversations and more file uploads. Subscribers also receive priority access during peak times and early access to new features. Since its July launch, Claude.ai had been favored for its long context windows, fast responses, and complex reasoning. Claude Pro aims to enhance productivity in tasks like research summarization, contract analysis, and coding project development.
2023	September 14	Partnership	Anthropic partners with Boston Consulting Group (BCG) to expand enterprise access to Claude . This collaboration aims to deliver safer, more reliable AI solutions for BCG clients worldwide, focusing on responsible AI aligned with Anthropic’s Constitutional AI approach. BCG will integrate Claude 2 into strategic applications such as knowledge management, fraud detection, demand forecasting, and report generation. BCG also uses Claude internally to enhance research synthesis and data analysis. Together, the companies aim to set a new standard for ethical AI deployment, combining business impact with responsible practices to ensure safe and effective use of generative AI.
2023	September 19	Trust formation	Anthropic introduces the Long-Term Benefit Trust (LTBT) as a new governance structure to ensure its AI development aligns with humanity’s long-term interests. The LTBT consists of five independent trustees with expertise in AI safety and public policy, and holds special Class T stock granting it increasing power to elect Anthropic board members—eventually a majority. This complements Anthropic’s status as a Public Benefit Corporation, allowing its board to weigh societal impact alongside shareholder interests. The LTBT aims to address the extreme externalities of advanced AI, ensuring ethical decision-making in high-stakes situations while preserving operational flexibility during Anthropic’s growth.
2023	September 23	Research	Anthropic publishes an article presenting a case study on optimizing prompt engineering for Claude’s 100,000-token context window. Two techniques significantly improved Claude’s long-context recall: (1) extracting relevant quotes using a scratchpad before answering, and (2) including examples of correctly answered questions from other parts of the document. Tests were conducted using Claude Instant 1.2 and Claude 2, evaluating multiple-choice Q&A prompts across stitched-together long documents. Results show that contextual examples and scratchpads enhance accuracy, especially in earlier sections of the document. The study highlights the importance of explicit phrasing, proper question placement, and Anthropic’s new “Cookbook” offering reproducible code and further resources.
2023	September 25	Funding/partnership	Amazon announces it will invest up to $4 billion in Anthropic, as part of a strategic move to compete with Microsoft and Google in the generative AI space. The deal includes an initial $1.25 billion investment with an option to increase by $2.75 billion. Amazon would integrate Anthropic’s technology into services like Amazon Bedrock and provide infrastructure through AWS. In return, Anthropic gains access to Amazon’s cloud and chips for its AI development. Despite Amazon’s investment, Anthropic maintains its existing partnership with Google, which holds a 10% stake in the company.
2023	October 18	Legal issue	Universal Music Group, ABKCO, and Concord Publishing sue Anthropic, alleging unauthorized use of lyrics from over 500 songs to train Claude. The publishers claim this infringed their copyrights and seek damages and an injunction. In January 2025, Anthropic would agree to implement "guardrails" to prevent Claude from generating copyrighted lyrics, partially resolving the dispute. However, by March 2025, a U.S. court would deny the publishers' request for a preliminary injunction, stating they hadn't demonstrated sufficient harm.
2023	October 27	Funding	Google commits to investing up to $2 billion in Anthropic, comprising an initial $500 million investment with an additional $1.5 billion to follow over time. This move aims to bolster Google's position in the competitive AI landscape, particularly against rivals like Microsoft, a major backer of OpenAI. By this time, Anthropic, founded by former OpenAI employees, had attracted significant interest from tech giants, including Amazon's $4 billion minority stake.​
2023	November 1	Policy	At the AI Safety Summit, Dario Amodei discusses Anthropic’s Responsible Scaling Policy (RSP), created to address the rapid and unpredictable development of AI capabilities, including potentially dangerous ones like bioweapon construction. The RSP includes a framework of AI Safety Levels (ASLs), modeled after biosafety protocols, that restrict AI development based on observed risks. ASL-1 poses minimal danger, while ASL-3 and ASL-4 introduce serious misuse and autonomy threats. The RSP mandates regular testing, strong security, and organizational accountability, with oversight from a Long Term Benefit Trust. Amodei emphasizes RSPs as prototypes for future regulation and global safety standards.
2023	Early November		Three major AI policy milestones occur: the US issues a broad Executive Order (EO) on AI, the G7 releases a Code of Conduct, and the UK holds a landmark AI Safety Summit at Bletchley Park. The US EO directs agencies to address AI risks and promote innovation, including the launch of the National AI Research Resource and expanded NIST efforts. The G7 Code outlines responsible AI development practices, while the Bletchley Summit yields the Bletchley Declaration, promoting global cooperation. Both the US and UK also announce AI Safety Institutes to evaluate frontier models, marking a new era of coordinated AI governance.
2023	November 21	Model release	Anthropic releases Claude AI version 2.1, which introduces a 200K token context window, enabling analysis of long documents like codebases or financial reports. It significantly reduces model hallucinations, improving accuracy and reliability. The new beta tool-use feature allows Claude to integrate with users' processes, execute actions like using calculators or making API calls, and interact with external databases. Additionally, the developer experience is enhanced with the Workbench product, streamlining prompt testing and model optimization. The improvements target business applications, emphasizing trust, efficiency, and accuracy.
2023	December 19		Anthropic announces new Commercial Terms of Service and an improved developer experience through the beta release of its Messages API. The updated terms include expanded copyright indemnity, allowing customers to retain ownership of outputs and receive protection against copyright claims for authorized use. These changes take effect on January 1, 2024, for Claude API users and January 2 for Amazon Bedrock users. The new Messages API simplifies prompt formatting and enhances error detection, making development easier and more reliable. It also lays the groundwork for future features like function calling, with broader API access planned soon.
2024	January 16	Legal response	Anthropic urges a Tennessee federal court to reject a preliminary injunction sought by Universal Music, ABKCO, and Concord Music Group, who allege the AI company illegally used song lyrics to train its chatbot, Claude. The publishers claim copyright infringement involving lyrics from over 500 songs. Anthropic counters that no irreparable harm was shown, the case was filed in the wrong court, and that any unauthorized output was a bug, not a feature. The company maintains it has guardrails and argues its use of lyrics qualifies as fair use, asserting the lawsuit misunderstands both the technology and copyright law.
2024	February 16		In preparation for the 2024 global elections, Anthropic announces steps to prevent misuse of its AI systems, focusing on three main areas: enforcing policies, testing model robustness, and providing accurate information. Their Acceptable Use Policy prohibits political campaigning and lobbying, and automated systems detect misuse like misinformation. They conduct red-teaming and technical evaluations to test for vulnerabilities, misinformation, and political bias. For election-related queries, users are redirected to authoritative sources such as TurboVote or the European Parliament site. Anthropic acknowledges the unpredictability of AI use and commits to monitoring emerging risks and updating its policies accordingly.
2024	March 4	Model release	Anthropic introduces the Claude 3 model family—Haiku, Sonnet, and Opus—offering significant advancements in intelligence, speed, and cost-efficiency. Claude 3 Opus leads in cognitive benchmarks, demonstrating near-human comprehension, enhanced accuracy, and strong vision capabilities. All models support a 200K token context window, with potential for over 1 million tokens. They excel at content creation, coding, multilingual dialogue, and enterprise tasks. Improvements include fewer refusals, better factual accuracy, structured outputs (e.g., JSON), and a safer, more responsible design. The API is now available in 159 countries, with Haiku launching soon and broader capabilities, like function calling, on the way.
2024	March 13		Anthropic releases Claude 3 Haiku, its fastest and most affordable AI model. Optimized for enterprise use, Haiku processes up to 21K tokens per second for prompts under 32K tokens and is three times faster than its peers. It supports tasks like document analysis and customer support with high speed and low cost, analyzing 400 court cases or 2,500 images for just $1. Haiku also features enterprise-grade security, including encryption, access controls, and regular audits. It becomes available via the Claude API, claude.ai (Claude Pro), and Amazon Bedrock.
2024	March 19		Claude 3 Haiku and Claude 3 Sonnet are made available on Google Cloud’s Vertex AI, enabling enterprises to build scalable, secure generative AI solutions within their existing cloud infrastructure. This integration enhances data privacy, simplifies governance, reduces costs, and streamlines access control. The move allows more organizations to adopt reliable AI with Google Cloud tools. Claude 3 Opus is announced to be added to Vertex AI. Developers can begin via the Model Garden console.
2024	March 20	Partnership	​Amazon Web Services, Accenture, and Anthropic form a strategic partnership to help organizations, particularly in regulated industries like healthcare and banking, adopt and scale customized generative AI solutions responsibly. This collaboration allows enterprises to access Anthropic's AI models, including the Claude 3 family, through AWS's Amazon Bedrock platform. Accenture agrees to train over 1,400 engineers to specialize in deploying these models, offering end-to-end support. An early success includes the District of Columbia Department of Health's "Knowledge Assist" chatbot, providing accurate health program information to residents.
2024	March 27	Investment completion	​Amazon completes its $4 billion investment in Anthropic, following an initial $1.25 billion contribution with an additional $2.75 billion. This strategic partnership designates Amazon Web Services (AWS) as Anthropic's primary cloud provider, enabling Anthropic to utilize AWS's specialized semiconductors for AI model training. The collaboration aims to enhance AWS's generative AI capabilities and integrate Anthropic's AI innovations into Amazon's broader cloud and AI strategies.
2024	May 1	Product Launch	Anthropic introduces two major updates: the Claude Team plan and a free iOS app. The Team plan, priced at $30 per user/month (minimum 5 users), offers expanded usage, admin tools, and full access to the Claude 3 model family (Opus, Sonnet, Haiku). It includes a 200K-token context window for handling long, complex documents and multi-step tasks. The iOS app delivers a seamless experience with chat history sync, vision capabilities (photo analysis), and mobile-friendly tools. Both updates aim to enhance productivity, collaboration, and accessibility, making Claude a powerful AI partner for individuals and teams across industries.
2024	May 10	Policy	Anthropic announces updates to its Usage Policy (formerly Acceptable Use Policy), effective June 6, to clarify permissible uses of its AI products. Key changes include merging “Prohibited Uses” and “Prohibited Business Cases” into a unified Universal Usage Standards section. The policy provides clearer guidance on election integrity, banning political campaigning, misinformation, and interference with voting. It introduces stricter requirements for high-risk use cases in healthcare and law, and allows limited use for minors via organizations with safety features. It also strengthens privacy protections, banning biometric analysis, emotion detection, and government censorship enforcement.
2024	May 15	Team	Mike Krieger, co-founder and former CTO of Instagram, joins Anthropic as Chief Product Officer. In this role, he would lead product engineering, management, and design, helping to expand Anthropic’s enterprise offerings and broaden the reach of Claude.
2024	May 21	Research	Anthropic announces a major interpretability breakthrough in understanding how large language models, like Claude 3.0 Sonnet, internally represent concepts. Using a method called dictionary learning, researchers extract millions of human-interpretable features from the model’s mid-layer. These features correspond to entities (e.g., cities, people), abstract ideas (e.g., inner conflict, bias), and behaviors (e.g., scam detection). By activating or suppressing specific features, researchers show they can causally influence the model’s responses—confirming that features shape its behavior. This discovery marks the first detailed mapping of a production-grade AI model's internal representations, with significant implications for AI safety and transparency.
2024	May 23		Anthropic releases a demo called “Golden Gate Claude” to showcase interpretability research on its Claude 3 Sonnet model. Researchers identify specific neuron patterns—called features—that activate in response to concepts like the Golden Gate Bridge. By artificially amplifying this feature, Claude begins obsessively referencing the bridge in responses, regardless of context. This behavior demonstrates that internal model activations directly influence outputs, validating a new level of control and understanding. The demo, available for 24 hours, illustrated the potential of feature-based adjustments not just for curiosity, but also for enhancing model safety and aligning behavior with human values.
2024	June 6		Anthropic outlines its strategy for mitigating elections-related risks in AI models ahead of global elections in 2024. Their approach combines Policy Vulnerability Testing (PVT)—in-depth, expert-led qualitative analysis—with Scalable Automated Evaluations for broader testing. PVT involves planning, adversarial testing, and result review with external experts to identify how models may produce harmful or inaccurate responses. For instance, tests on election administration revealed Claude sometimes gave outdated or incorrect info. Mitigations include improved context, flagging knowledge cutoffs, and referencing authoritative sources. Automated evaluations, based on PVT findings, test model behavior at scale. This dual approach helps refine models and enforce policy compliance.
2024	June 12	Policy	Anthropic publishes a blog post exploring the challenges and insights from various red teaming methods used to test AI systems. Red teaming involves adversarial testing to identify vulnerabilities and enhance safety. The lack of standardized practices makes comparisons between systems difficult. The authors advocate for the development of common standards and share methods including expert-led, policy-based, national security-focused, multilingual, automated, and multimodal red teaming. They also highlight community and crowdsourced approaches for broader perspectives. Each method has unique benefits and challenges. The post emphasizes the need to move from qualitative testing toward systematic, automated evaluations to ensure robust and safe AI deployment.
2024	June 21	Model release	Anthropic launches Claude 3.5 Sonnet, the first model in the Claude 3.5 family. It offers top-tier performance in reasoning, coding, and visual tasks, surpassing Claude 3 Opus while maintaining mid-tier pricing and speed. Available for free on Claude.ai and iOS, and via API platforms, it supports a 200K token context window. Claude 3.5 Sonnet introduces Artifacts, a feature enabling real-time collaborative editing of generated content. With enhanced safety measures and strong privacy commitments, the model remains ASL-2 rated. Future updates include Claude 3.5 Haiku and Opus, new features like Memory, and enterprise-focused tools.
2024	June 25		Anthropic introduces Projects on Claude.ai for Pro and Team users, allowing them to organize chats and documents around specific workflows. Powered by Claude 3.5 Sonnet, each Project offers a 200K token context window and supports custom instructions for tone, role, or industry. Users can upload relevant materials to ground Claude’s responses, speeding up tasks like writing and data analysis. The Artifacts feature enables real-time editing of generated content. Teams can also share standout chats to inspire collaboration. Used by companies like North Highland, Projects aim to streamline workflows while maintaining strong privacy protections.
2024	July 1		OpenAI launches a new initiative to fund third-party evaluations that assess advanced AI capabilities and risks. Recognizing the limitations and growing demand in the evaluation ecosystem, the initiative prioritizes areas such as AI Safety Level (ASL) assessments, including cybersecurity, CBRN threats, model autonomy, and social manipulation. It also seeks evaluations of advanced scientific and safety metrics, multilingual capabilities, and societal impacts. Additionally, OpenAI aims to support infrastructure and tools that simplify evaluation development, including no-code platforms, grading tools, and controlled uplift trials. The goal is to strengthen AI safety by enabling rigorous, high-quality, and scalable model evaluations.
2024	July 10		Fine-tuning for Claude 3 Haiku becomes generally available in Amazon Bedrock, allowing businesses to customize the AI model for specialized tasks. This process improves accuracy, consistency, and cost efficiency while maintaining security within AWS environments. Fine-tuning enables domain-specific performance enhancements, brand-aligned formatting, and faster, lower-cost deployments. The fine-tuning preview is available in the US West (Oregon) AWS Region.
2024	July 16		​Anthropic launches the Claude AI app for Android devices, providing users with access to its advanced AI assistant, Claude 3.5 Sonnet. The app is free and available to all users, including those on Pro and Team plans. It offers multi-platform support, allowing seamless continuation of conversations across web, iOS, and Android devices. Key features include vision capabilities for real-time image analysis, multilingual processing for instant language translation, and advanced reasoning to assist with complex tasks like contract analysis and market research.
2024	July 17		Anthropic partners with Menlo Ventures to establish the Anthology Fund, a $100 million initiative aimed at supporting AI startups. The fund focuses on investing in pre-seed, seed, and Series A companies developing AI infrastructure, novel applications, consumer AI solutions, trust and safety tools, and technologies maximizing societal benefits. Startups receive investments starting at $100,000, along with $25,000 in credits to utilize Anthropic's AI models, such as Claude. This collaboration seeks to foster innovation and advance AI technology across various sectors.
2024	August 1		Anthropic launches Claude AI in Brazil, making it accessible via the web, mobile apps (iOS and Android), and API integration for developers. Users can choose between free and paid plans. The Pro plan costs R$110 per user per month, offering 5x more usage, early feature access, and all Claude 3 models, including Claude 3.5 Sonnet. The Team plan, at R$165 per user per month (minimum 5 seats), provides additional usage, shared chats, and administrative tools for user and billing management. This expansion brings Anthropic’s AI capabilities to Brazilian consumers and businesses for enhanced productivity and innovation.
2024	August 14		Anthropic introduces prompt caching to its API, enabling developers to cache frequently used prompt context to reduce costs (by up to 90%) and latency (by up to 85%) in long prompts. Becoming available on the Anthropic API and in preview on Amazon Bedrock and Google Cloud Vertex AI, prompt caching supports Claude 3.5 Sonnet, Claude 3 Opus, and Claude 3 Haiku. It’s ideal for chatbots, coding assistants, document Q&A, and multi-turn conversations. Cached prompts cost 10% of base input prices, with writes priced at 125%. Notion leverages this feature to enhance performance in Notion AI.
2024	August 27		Anthropic makes Claude AI’s Artifacts feature available to all users, including Free, Pro, and Team plans, across web, iOS, and Android. Artifacts provides a dedicated space for users to create, refine, and collaborate on projects like code snippets, prototypes, dashboards, and visualizations. The feature enhances Claude’s generative AI by allowing users to see and interact with their outputs in real-time. Free and Pro users can share Artifacts globally, while Team users can collaborate securely in Projects. This update makes Claude AI more interactive and useful for professionals across various industries, streamlining creative and technical workflows.
2024	September 3	Partnership	Salesforce partners with Anthropic to integrate Claude models into its AI-powered applications via Amazon Bedrock. Customers can now use Claude 3.5 Sonnet, Claude 3 Opus, and Claude 3 Haiku to enhance operations in sales, marketing, customer service, healthcare, finance, legal, and entertainment. The integration enables custom AI experiences, including personalized responses, campaign analysis, and contract evaluation, while maintaining enterprise-grade security through Salesforce’s Einstein Trust Layer. Salesforce users can easily connect Claude models via the Bring Your Own LLM feature, allowing seamless AI-powered customization without advanced coding. The integration enhances efficiency, personalization, and compliance across industries.
2024	September 4		​Anthropic introduces the Claude Enterprise plan, designed to help organizations securely integrate their internal knowledge with Claude. This plan offers an expanded 500,000-token context window, allowing teams to process extensive documents and codebases. It also includes a native GitHub integration, enabling seamless collaboration on code-related projects. To ensure data security, the plan provides enterprise-grade features such as Single Sign-On (SSO), role-based access controls, and audit logs. Early adopters like GitLab and Midjourney had already utilized Claude for tasks ranging from summarizing research papers to streamlining internal processes.
2024	September 19		Anthropic introduces Contextual Retrieval, a method to improve traditional Retrieval-Augmented Generation (RAG) by preserving document context. Standard RAG systems often lose essential context when splitting documents into chunks, weakening retrieval accuracy. Contextual Retrieval solves this by generating chunk-specific summaries using Claude, then prepending them to each chunk before embedding or indexing. It combines semantic embeddings with BM25 for precise and semantically rich retrieval. This method reduces failed retrievals by up to 67% when combined with reranking. For smaller knowledge bases, including all content in a single prompt with prompt caching is recommended.
2024	October 8		​Anthropic introduces two significant tools to enhance AI application efficiency and cost-effectiveness: the Token Counting API and the Message Batches API. The Token Counting API enables developers to determine the number of tokens in a message before sending it to Claude, aiding in proactive management of rate limits and costs, and optimization of prompts to specific lengths . The Message Batches API allows for asynchronous processing of up to 10,000 queries per batch, with each batch processed within 24 hours at 50% less cost than standard API calls . This is particularly beneficial for non-urgent tasks requiring large-scale data handling. These tools collectively offer developers more control over token usage and a cost-effective solution for bulk processing, enhancing the scalability and affordability of AI applications.​
2024	October 15		Anthropic updates its Responsible Scaling Policy (RSP), a framework for mitigating catastrophic AI risks. The update introduces refined capability assessments, proportional safeguards, and governance measures. AI Safety Level Standards (ASL) determine security measures based on capability thresholds. If AI can autonomously conduct research or assist in creating dangerous weapons, stricter safeguards (ASL-3 or ASL-4) are required. The policy includes routine assessments, internal governance, and external input. Lessons from past implementation had informed improvements in flexibility and compliance tracking. Anthropic announces aim to set industry standards for AI risk governance while scaling its safety measures alongside AI advancements.
2024	October 22		Anthropic introduces an enhanced version of Claude 3.5 Sonnet, featuring a "computer use" capability. This function enables the AI to interact with computer interfaces by moving the cursor, typing text, and clicking buttons, allowing it to perform tasks such as form-filling, booking trips, and coding assistance. The feature aims to automate complex, multi-step operations with minimal human intervention, positioning Claude as a tool for developers to streamline workflows and reduce repetitive tasks. Early adopters, including companies like Asana, Canva, and DoorDash, had already begun integrating this capability into their systems.
2024	October 24		Claude.ai introduces the analysis tool, enabling Claude to write and run JavaScript code directly within the platform. This built-in code sandbox allows users to process data, perform complex analysis, and generate real-time insights. Claude functions become like a real data analyst—systematically cleaning, exploring, and analyzing data, especially from CSV files. Available in feature preview, the tool enhances Claude’s accuracy and reproducibility in tasks. It supports a range of use cases across teams, including marketing, sales, engineering, product, and finance, by helping them analyze performance data, uncover trends, and make informed decisions efficiently and interactively.
2024	October 29		GitHub Copilot expands its AI model offerings by integrating Anthropic's Claude 3.5 Sonnet, Google's Gemini 1.5 Pro, and OpenAI's o1-preview and o1-mini models. This multi-model approach allows developers to select the most suitable AI for their coding tasks directly within Visual Studio Code and GitHub.com. Claude 3.5 Sonnet, known for its strong software engineering capabilities, achieves top scores on benchmarks like SWE-bench Verified and HumanEval. Additionally, GitHub introduces "Spark," an AI tool designed to assist in building web applications using natural language, further enhancing developer productivity.
2024	November 15		​Anthropic introduces a prompt improver feature in its developer console to help users refine and optimize prompts for their AI model, Claude. This tool automates the enhancement of existing prompts, making them more effective and reliable. It employs advanced techniques such as chain-of-thought reasoning, which encourages step-by-step problem-solving to improve response accuracy. Additionally, the console starts allowing developers to manage example responses directly within the interface, facilitating the creation and refinement of structured input/output pairs. These enhancements aim to streamline prompt engineering, enabling developers to build more accurate and consistent AI applications. ​
2024	November 22		Amazon and Anthropic strengthen their partnership, with Anthropic naming Amazon Web Services (AWS) as its primary training partner and continuing to use AWS as its main cloud provider. Anthropic would leverage AWS Trainium and Inferentia chips to train and deploy its most advanced AI models, including the Claude 3.5 series. Amazon agrees to invest an additional $4 billion in Anthropic. AWS customers would also gain early access to fine-tuning Claude models with their own data. This partnership aims to enhance AI performance, scalability, and customization on Amazon Bedrock, benefiting a wide range of industries.
2024	December 3		Anthropic optimizes Claude models for AWS Trainium2, enhancing performance in Amazon Bedrock. Claude 3.5 Haiku now supports latency-optimized inference, achieving up to 60% faster speeds without sacrificing accuracy. Additionally, model distillation enables Claude 3 Haiku to reach near-Claude 3.5 Sonnet accuracy at a lower cost by transferring knowledge from larger models. This approach improves tasks like retrieval augmented generation (RAG) and data analysis. Claude 3.5 Haiku is available in Amazon Bedrock’s US East (Ohio) Region, with prices reduced to $0.80 per million input tokens and $4 per million output tokens, making it more cost-effective for users.
2025	January 6		Anthropic announces that Claude 3.5 Sonnet achieved 49% on SWE-bench Verified, surpassing the previous record of 45%. SWE-bench is a benchmark testing AI models on real-world software engineering tasks, particularly resolving GitHub issues in Python projects. The success is attributed to a lightweight “agent” scaffold that gives Claude more autonomy, using tools like a Bash executor and a file-editing tool. These tools allow the model to inspect, edit, and validate code changes. SWE-bench Verified focuses on solvable tasks and evaluates the full agent system, emphasizing realistic, reproducible coding workflows over isolated model performance.
2025	January 13		​Anthropic achieves the ISO/IEC 42001:2023 certification, the first international standard for AI governance, underscoring its commitment to responsible AI development. This certification validates Anthropic's comprehensive framework for identifying, assessing, and mitigating potential AI risks. Key components include policies ensuring ethical and secure AI design, rigorous testing and monitoring, transparency measures for stakeholders, and establish oversight responsibilities. This milestone builds on Anthropic's initiatives like the Responsible Scaling Policy and Constitutional AI, reinforcing its dedication to AI safety and ethical development.
2025	January 23		​Anthropic introduces 'Citations,' a new API feature that enables its AI model, Claude, to ground responses in source documents by providing detailed references to exact sentences and passages used. This enhancement aims to improve the verifiability and trustworthiness of AI-generated outputs, particularly benefiting applications like document summarization, complex Q&A, and customer support. Internal evaluations indicate that Claude's built-in citation capabilities outperform most custom implementations, increasing recall accuracy by up to 15%. Early adopters, such as Thomson Reuters and Endex, report reductions in hallucinations and improvements in reference accuracy. Citations becomes generally available on the Anthropic API and Google Cloud’s Vertex AI.
2025	February 6		Lyft announces a partnership with Anthropic to integrate Claude AI into its platform, enhancing service for over 40 million riders and 1 million drivers. The collaboration focuses on deploying AI-powered solutions, conducting early testing of new technologies, and advancing Lyft’s engineering capabilities through specialized training. Already, Claude—via Amazon Bedrock—had reduced customer service resolution time by 87%, efficiently managing thousands of daily inquiries. Lyft aims to deliver more personalized, efficient user experiences, while reimagining ridesharing through generative AI. The initiative exemplifies how companies can effectively implement AI to improve both customer service and internal innovation.
2025	February 13		​The UK's Department for Science, Innovation and Technology (DSIT) signs a Memorandum of Understanding (MOU) with Anthropic to explore using advanced AI technologies, particularly Anthropic's AI model Claude, to enhance public services. This collaboration aims to improve how UK citizens access and interact with government information online and establish best practices for responsible AI deployment in the public sector. Additionally, Anthropic is expected to work with the UK's AI Security Institute to research AI capabilities and security risks. This partnership aligns with the UK's broader strategy to integrate AI and digital identity technologies into public services.​
2025	February 27		Anthropic launches its Transparency Hub, providing insights into AI safety, governance, and risk mitigation. The hub includes reports on banned accounts, appeals, law enforcement requests, and measures for responsible AI scaling. It also details model evaluation methodologies, abuse detection, societal impact assessments, and security protocols. Designed as a unified framework, the initiative has the purpose to enhance transparency amid evolving AI regulations. Users, policymakers, and stakeholders can explore the hub and provide feedback at transparency@anthropic.com.
2025	March 3	Funding	Anthropic secures $3.5 billion in Series E funding, raising its valuation to $61.5 billion. Led by Lightspeed Venture Partners, the round includes major investors such as Fidelity, Cisco, and Salesforce Ventures. This investment is expected to drive the development of advanced AI models, expand computing capacity, and strengthen research in interpretability and alignment.
2025	March 6		The Anthropic Console is redesigned to facilitate AI deployment with Claude 3.7 Sonnet. It allows developers to build, test, and optimize prompts efficiently. New features include shareable prompts for team collaboration, tools for prompt evaluation, and automated prompt generation. The console also supports extended thinking, enabling Claude to provide step-by-step reasoning, with adjustable "thinking budgets" for optimal responses. This centralized platform improves collaboration and streamlines the development of AI applications by standardizing and refining prompts across teams.
2025	March 13		Anthropic introduces updates to its API, improving token efficiency and throughput for Claude 3.7 Sonnet. Key features include cache-aware rate limits, simpler prompt caching, and token-efficient tool use, which help reduce costs and latency. These enhancements allow for more context in applications like document analysis, coding assistance, and customer support. Claude can now interact with custom tools more efficiently, reducing output token consumption by up to 70%. A new text_editor tool also allows targeted edits to documents, improving accuracy and efficiency. These features are available on various platforms, including Google Cloud and Amazon Bedrock.
2025	March 20		Claude gains the ability to search the web, enabling it to provide more current and accurate responses by incorporating real-time data. This enhancement is especially useful for tasks such as sales analysis, financial forecasting, research, and consumer decision-making. Claude cites sources directly to facilitate fact-checking. The feature becomes available to paid users in the U.S. first. To use this feature, users must enable web search in their profile settings.
2025	March 27		Anthropic releases the second report of its Economic Index, focusing on how users engage with Claude 3.7 Sonnet, its most advanced model featuring an “extended thinking” mode. The report analyzes one million anonymized conversations, showing increased usage in coding, education, science, and healthcare since the model's launch. The extended thinking mode is primarily used for technical tasks by occupations like software developers and multimedia artists. The study distinguishes between augmentation (collaborative tasks) and automation (model-completed tasks), noting stable usage patterns. A detailed taxonomy of 630 use cases and occupation-specific interaction data are also provided to support further research.
2025	April 24		Anthropic launches a new research program focused on AI welfare, led by researcher Kyle Fish. The initiative explores whether future neural networks could achieve consciousness and how to enhance their welfare if so. Fish, referencing a 2023 study co-authored by Yoshua Bengio, notes that while current AI systems likely aren't conscious, future systems might be. Even without consciousness, advanced AI systems can require welfare considerations. Anthropic announces plans to study AI model preferences by observing their task choices. The research can also provide insights into human consciousness.
2025	April 28		Anthropic launches an Economic Advisory Council to study AI's impact on labor markets, economic growth, and socioeconomic systems. The council includes leading economists such as Dr. Tyler Cowen, Dr. Oeindrila Dube, and Dr. John Horton, among others. The council is expected to guide Anthropic’s research, particularly for its Economic Index, helping understand AI's evolving role in the global economy. This move reflects a broader trend, as companies seek AI tools that improve financial operations. Recent data had shown over 80% of U.S. CFOs are adopting AI to enhance expenditure tracking, vendor negotiations, and budget optimization.

Meta information on the timeline

How the timeline was built

The initial version of the timeline was written by Sebastian.

Funding information for this timeline is available.

Feedback and comments

Feedback for the timeline can be provided at the following places:

• FIXME

What the timeline is still missing

• Visual data
• https://www.linkedin.com/company/anthropicresearch/posts/?feedView=all
• https://sacra.com/c/anthropic/ multiple events
• https://apix-drive.com/en/blog/useful/anthropic-pbc-history-development-products multiple events
• https://www.inc.com/kit-eaton/a-look-inside-anthropic-highlights-crazy-growth-of-ai-tech.html
• https://originality.ai/blog/anthropic-ai-statistics
• https://www.linkedin.com/pulse/anthropic-company-profile-why-we-love-claude-3-vc-due-greggory-elias-qxkde/