{"id":"sam-mccandlish","hidden":false,"ipfs":"QmRttPCxAdNmSqovgLDPSrGW8mjTmp4x7MZwrg7kMGuae4","language":"en","transactionHash":"0xddd841ffc98512ad9381abcc3b0ecd53c5143a25b9c0388148acc3b17d7afb1d","created":"2026-07-03T03:58:00.015Z","updated":"2026-09-24T14:57:37.800Z","title":"Sam McCandlish","summary":"Sam McCandlish is an Anthropic co-founder and Chief Architect whose research has addressed neural-network scaling laws, large-batch training, and AI safety.","content":"**Sam McCandlish** is the co-founder and Chief Architect of Anthropic, an artificial intelligence safety and research company. He previously served as its Chief Technology Officer until October 2025.[[6]](#cite-id-8tjhp0lud1)[[7]](#cite-id-8hzmw5xzkp)[[5]](#cite-id-57qci7qtkh)[[4]](#cite-id-qTT4w6I4rgRbT7ME) Before Anthropic, McCandlish worked with OpenAI, contributing to its research on neural network performance and safety mechanisms.[[1]](#cite-id-ExDmLJjaG5WF9Zdu)[[2]](#cite-id-BjYCfDi2bf1TTBEH) \n\n## Early Life & Education\n\nSam McCandlish pursued his academic interests initially through a B.S. and M.S. in Mathematics and Physics at Brandeis University, followed by a Ph.D. in Theoretical Physics from Stanford University. His doctoral research delved into the complexities of quantum gravity and tensor networks, equipping him with a foundation in systematic modeling that he later applied to machine learning and AI.[[1]](#cite-id-ExDmLJjaG5WF9Zdu)\n\n## Career\n\n\n### OpenAI\nSam McCandlish was a Postdoctoral Fellow at Boston University from January 2017 to January 2018, where he was affiliated with the Simons Bootstrap Collaboration.[[2]](#cite-id-BjYCfDi2bf1TTBEH)\n\nAfter an AI Safety Fellowship, McCandlish joined OpenAI as a Research Lead from May 2018 to December 2020. He built a team to study the science and scaling of machine learning, and led early research on [Codex](https://iq.wiki/wiki/codex).[[2]](#cite-id-BjYCfDi2bf1TTBEH)[[8]](#cite-id-rv9fumiwl4)\n\nHis research examined the performance and scaling behavior of neural networks, with a focus on language models and large-scale training methods. His work included studies on scaling laws for neural language models and large-batch training. The gradient noise scale measures the signal-to-noise ratio of stochastic training gradients and can predict the largest useful batch size, helping estimate when added data parallelism yields diminishing returns. Its empirical study covered image recognition, language modeling, [Atari](https://iq.wiki/wiki/atari), and Dota.[[11]](#cite-id-0wadwn3fo0)[[12]](#cite-id-f9jjw94uvm)\n\nPublications and projects from this period include Language Models Are Few-Shot Learners, Scaling Laws for Language Models, Scaling Laws for Neural Language Models, Scaling Laws for Transformers, An Empirical Model of Large-Batch Training, as well as OpenAI's Science of AI and AI and Efficiency projects.[[2]](#cite-id-BjYCfDi2bf1TTBEH)[[11]](#cite-id-0wadwn3fo0)[[12]](#cite-id-f9jjw94uvm)[[13]](#cite-id-a4yn3pjit9)\n\n### Anthropic\nIn January 2021, McCandlish co-founded Anthropic with former OpenAI colleagues, including Dario and [Daniela Amodei](https://iq.wiki/wiki/daniela-amodei). The company conducts research on artificial intelligence with an emphasis on AI safety and interpretability and, as of February 2026, was valued by private investors at $380 billion, with major partnerships with Alphabet and Amazon.[[3]](#cite-id-1SAzxmGUEPOqOv5Q)​\n\nAt Anthropic, McCandlish has served as Chief Scientist and Chief Technology Officer and, following an October 2025 restructuring, moved into the role of Chief Architect. He remains Co-Founder and Chief Architect, focusing on large-scale model training and leading pretraining, research productivity, and reinforcement-learning infrastructure.[[6]](#cite-id-8tjhp0lud1)[[8]](#cite-id-rv9fumiwl4) His work has included the management of large-scale model training and has contributed to the development of the Claude family of language models and Anthropic's Constitutional AI methodology, which uses AI-generated feedback as part of the model alignment process rather than relying solely on human preference data.\n\nHis research at Anthropic has focused on scaling laws, AI safety, and mechanistic interpretability. It has also included work related to Constitutional AI and model alignment methods. In early 2026, Anthropic restricted its internal Mythos model to controlled environments after internal evaluations found that it could autonomously identify and exploit system vulnerabilities, characterizing the move as a safety-motivated limitation on model use.[[5]](#cite-id-57qci7qtkh) On November 9, 2026, Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2 as versions of the same underlying model. Fable is available to all users with stronger cybersecurity and biology safeguards, while Mythos is available to vetted organizations through Anthropic’s Cyber Verification Program and Life Sciences Verification Program. Claude Enterprise customers can access Claude Security, which uses Mythos 5.2.[[9]](#cite-id-92yko8hp26)\n\nMcCandlish and other Anthropic co-founders pledged to dedicate 80% of their personal equity in Anthropic to philanthropic causes.[[10]](#cite-id-cm6soaoy0a)","recentActivity":"{\"items\":[{\"id\":\"8006e425-d6ff-4e5d-a9dc-54793ff66249\",\"title\":\"Sam McCandlish\",\"description\":\"Sam McCandlish is a leading figure in AI research, co-founding Anthropic to drive AI safety innovations. A former OpenAI researcher, he's renowned for scaling law contributions in machine learning.\",\"timestamp\":\"2026-07-06T23:12:34.044Z\",\"category\":\"People in crypto\",\"status\":{\"icon\":\"RiGlobalLine\",\"label\":\"Wiki Updated\",\"iconClassName\":\"text-green-500\"},\"user\":{\"name\":\"0x8AF7a19a26d8FBC48dEfB35AEfb15Ec8c407f889\",\"address\":\"0x1E23b34d3106F0C1c74D17f2Cd0F65cdb039b138\"},\"button\":{\"label\":\"View Summary\",\"icon\":\"RiFileTextLine\"},\"summarySections\":[{\"title\":\"Title\",\"subtitle\":\"Title text was changed\",\"variant\":\"modified\",\"changeCount\":1,\"changes\":[\"Changed title from \\\"people\\\" to \\\"People in crypto\\\"\"]},{\"title\":\"IDs (metadata)\",\"subtitle\":\"Multiple id entries were removed and replaced/added; some added ids include multiDateStart/multiDateEnd set to null\",\"variant\":\"modified\",\"changeCount\":8,\"changes\":[\"Removed id \\\"0ea44dcd-d709-46be-8990-546b09f8bcfc\\\"\",\"Added id \\\"424c1488-0a16-44b8-ba55-879f5158d5bc\\\"\",\"Removed id \\\"ebb7e739-3b58-417c-a86c-b668fb1ace05\\\"\",\"Added id \\\"45a88c5e-3415-4e8c-b875-25776213ed47\\\" with multiDateStart: null and multiDateEnd: null\",\"Removed id \\\"fc501154-5c9a-499f-8bf7-2ce6d883d2a3\\\"\",\"Added id \\\"ed2bb5a6-7607-4f86-b890-74e792c84797\\\" with multiDateStart: null and multiDateEnd: null\",\"Removed id \\\"Qmayte5GeFdSDLutVDpVybdiepnNqtwBsRh5HMnSjSVTW6\\\"\",\"Added id \\\"QmeDB5wE9391r1tzuhJ7BYs3xDcAZLQUuZhiw6V5Dj4vW7\\\"\"]},{\"title\":\"Speakers\",\"subtitle\":\"The speakers field was removed\",\"variant\":\"removed\",\"changeCount\":1,\"changes\":[\"Removed \\\"speakers\\\" array (was [])\"]},{\"title\":\"Images\",\"subtitle\":\"Image object fields changed: caption and thumbnail keys were removed; size and type keys were added (null)\",\"variant\":\"modified\",\"changeCount\":2,\"changes\":[\"Image with thumbnail \\\"QmVfCsXmTmvdymspp5eC6eHLKssKnusCxeu1bi2BFbiQbL\\\": removed \\\"caption\\\" (empty) and removed \\\"thumbnail\\\" \\\"QmVfCsXmTmvdymspp5eC6eHLKssKnusCxeu1bi2BFbiQbL\\\"; added \\\"size\\\": null and \\\"type\\\": null\",\"Image with thumbnail \\\"QmTr22iRJcXqX892rDD46JVz1nbTpEArhnG77vCV7bEddP\\\": removed \\\"caption\\\" (empty) and removed \\\"thumbnail\\\" \\\"QmTr22iRJcXqX892rDD46JVz1nbTpEArhnG77vCV7bEddP\\\"; added \\\"size\\\": null and \\\"type\\\": null\"]},{\"title\":\"wiki_score\",\"subtitle\":\"A wiki_score metadata entry was added\",\"variant\":\"added\",\"changeCount\":1,\"changes\":[\"Added id \\\"wiki_score\\\" with value \\\"72\\\"\"]}]}]}","operator":{"id":"0x1e23b34d3106f0c1c74d17f2cd0f65cdb039b138"},"categories":[{"id":"people","title":"People in crypto"}],"tags":[{"id":"Founder"},{"id":"Researcher"}],"images":[{"id":"QmeDB5wE9391r1tzuhJ7BYs3xDcAZLQUuZhiw6V5Dj4vW7","type":"image/jpeg, image/png"}],"media":[{"name":"30874603.jpeg","id":"QmVfCsXmTmvdymspp5eC6eHLKssKnusCxeu1bi2BFbiQbL","size":null,"type":null,"source":"IPFS_IMG"},{"name":"1743076065656.jpeg","id":"QmTr22iRJcXqX892rDD46JVz1nbTpEArhnG77vCV7bEddP","size":null,"type":null,"source":"IPFS_IMG"}],"linkedWikis":{"founders":[],"blockchains":[],"speakers":[]},"events":[{"title":"Joined Boston University as a Postdoctoral Fellow","date":"2017-01-01","type":"DEFAULT","description":"Began a postdoctoral appointment affiliated with the Simons Bootstrap Collaboration.","link":"https://www.linkedin.com/in/sam-mccandlish","country":null,"continent":null,"multiDateStart":null,"multiDateEnd":null,"id":"5149817a-c43c-43fc-8bbb-ede5c8725cee"},{"title":"Joined OpenAI as Research Lead","date":"2018-05-01","type":"DEFAULT","description":"After an AI Safety Fellowship, led a team studying the science and scaling of machine learning.","link":"https://www.linkedin.com/in/sam-mccandlish","country":null,"continent":null,"multiDateStart":null,"multiDateEnd":null,"id":"c5073c15-44fe-4c4e-a895-9bdeeb927e72"},{"title":"Published An Empirical Model of Large-Batch Training","date":"2018-12-01","type":"DEFAULT","description":"Co-authored the paper introducing an empirical use for the gradient noise scale in estimating useful batch sizes.","link":"https://arxiv.org/abs/1812.06162","country":null,"continent":null,"multiDateStart":null,"multiDateEnd":null,"id":"0bf19b51-89d9-4364-9e73-1e38a8650b67"},{"title":"Left OpenAI","date":"2020-12-01","type":"DEFAULT","description":"Ended his tenure as OpenAI Research Lead.","link":"https://www.linkedin.com/in/sam-mccandlish","country":null,"continent":null,"multiDateStart":null,"multiDateEnd":null,"id":"26fa082f-1b52-4294-af53-53d2ff60ab7c"},{"title":"Co-founded Anthropic","date":"2021-01-01","type":"DEFAULT","description":"Joined former OpenAI colleagues in founding the AI safety and research company.","link":"https://www.linkedin.com/in/sam-mccandlish","country":null,"continent":null,"multiDateStart":null,"multiDateEnd":null,"id":"2bb4649c-4e52-453a-a7ad-1b5fbad2a938"},{"title":"Became Anthropic Chief Architect","date":"2025-10-01","type":"DEFAULT","description":"Moved from the CTO role to Chief Architect as Rahul Patil joined Anthropic as CTO.","link":"https://techcrunch.com/2025/10/02/anthropic-hires-new-cto-with-focus-on-ai-infrastructure/","country":null,"continent":null,"multiDateStart":null,"multiDateEnd":null,"id":"b390c3fa-1ff0-473c-98a1-d4088f1c1153"}],"founderWikis":[],"blockchainWikis":[],"metadata":[{"id":"linkedin_profile","value":"https://www.linkedin.com/in/sam-mccandlish/"},{"id":"coingecko_profile","value":"https://www.longtermwiki.com/wiki/E1259"},{"id":"references","value":"[{\"id\":\"ExDmLJjaG5WF9Zdu\",\"url\":\"https://www.longtermwiki.com/wiki/E1259\",\"description\":\"Longterm Wiki\",\"timestamp\":1783050215379},{\"id\":\"BjYCfDi2bf1TTBEH\",\"description\":\"LinkedIn: Sam Mccandlish\",\"timestamp\":1783050357542,\"url\":\"https://www.linkedin.com/in/sam-mccandlish/\"},{\"id\":\"1SAzxmGUEPOqOv5Q\",\"description\":\"PROFILE\\nSam McCandlish\\n\",\"timestamp\":1783050403111,\"url\":\"https://www.forbes.com/profile/sam-mccandlish/\"},{\"id\":\"qTT4w6I4rgRbT7ME\",\"description\":\"Samuel McCandlish is Co-Founder & Chief Architect at Anthropic, an artificial intelligence safety and research company, bringing expertise backed by a Ph.D. \",\"timestamp\":1783050528715,\"url\":\"https://www.zoominfo.com/p/Samuel-Mccandlish/1494553260\"},{\"id\":\"57qci7qtkh\",\"url\":\"https://www.contextjamming.com/founder-files/sam-mccandlish\",\"description\":\"Context Jamming\",\"timestamp\":1790261579404},{\"id\":\"8tjhp0lud1\",\"url\":\"https://www.anthropic.com/company/leadership\",\"description\":\"Anthropic leadership\",\"timestamp\":1795189875635},{\"id\":\"8hzmw5xzkp\",\"url\":\"https://www.observer.com/2026/04/anthropic-top-executives-after-labs-launch/\",\"description\":\"Observer\",\"timestamp\":1795189875635},{\"id\":\"rv9fumiwl4\",\"url\":\"https://www.cnbc.com/2026/11/14/meet-the-seven-co-founders-of-anthropic.html\",\"description\":\"CNBC\",\"timestamp\":1795189875635},{\"id\":\"92yko8hp26\",\"url\":\"https://www.anthropic.com/claude-fable-and-mythos-5-2\",\"description\":\"Anthropic announcement\",\"timestamp\":1795189875635},{\"id\":\"cm6soaoy0a\",\"url\":\"https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm\",\"description\":\"Anthropic Form S-1\",\"timestamp\":1795189875635},{\"id\":\"0wadwn3fo0\",\"url\":\"https://arxiv.org/abs/1812.06162\",\"description\":\"An Empirical Model of Large-Batch Training\",\"timestamp\":1795190205090},{\"id\":\"f9jjw94uvm\",\"url\":\"https://openai.com/index/how-ai-training-scales/\",\"description\":\"How AI Training Scales\",\"timestamp\":1795190205090},{\"id\":\"a4yn3pjit9\",\"url\":\"https://openai.com/index/ai-and-efficiency/\",\"description\":\"AI and Efficiency\",\"timestamp\":1795190205090}]"},{"id":"previous_cid","value":"QmRttPCxAdNmSqovgLDPSrGW8mjTmp4x7MZwrg7kMGuae4"},{"id":"commit-message","value":"Expand Sam McCandlish summary by 27 words and add timeline"},{"id":"previous_cid","value":"QmRttPCxAdNmSqovgLDPSrGW8mjTmp4x7MZwrg7kMGuae4"}],"user":{"id":"0x8af7a19a26d8fbc48defb35aefb15ec8c407f889"},"author":{"id":"0x1e23b34d3106f0c1c74d17f2cd0f65cdb039b138"},"views":624}