NYTech
People
Personfounder

Dan Shiebler

Dan Shiebler is a machine learning researcher and executive in the New York technology sector, known for his work applying machine learning to cybersecurity and, since 2025, for co-founding Artemis, an AI-native security platform [1][2][3]. He holds a Ph.D. in artificial intelligence theory from the University of Oxford, with a dissertation on applications of category theory to machine learning, and undergraduate degrees from Brown University [3][6]. Before founding Artemis, he spent over three years as Head of AI/ML at Abnormal AI, where he held ultimate responsibility for the detection efficacy of the company's email and SaaS-account security platform, building on anti-spam and anti-abuse work he had previously done at Twitter [4][7]. At Abnormal, he described the core of the detection system as "aggregate signals," engineered features summarizing historical behavior at the level of individual senders, recipients, and network infrastructure, on top of which the company layered a range of models from simple heuristics and Bayesian-style probabilistic rules through logistic regression and deep-and-cross neural networks up to large language models, chosen according to the cost, speed, and accuracy tradeoffs required for a given threat [7]. He has framed this work as inherently adversarial, since attackers actively try to disguise phishing messages and account takeovers as normal business activity, requiring models that exploit information the attacker cannot easily anticipate [7].

Earlier in his career, Shiebler worked at Twitter for about five years, first as a staff machine learning researcher on the central Cortex platform team and later as an engineering manager in Revenue Science, where he worked on early applications of large language models and vector databases alongside building shared training and deployment infrastructure used across product teams [4][5]. During this period he spoke publicly about "resilient machine learning," arguing that production systems should be engineered to keep functioning when upstream data services fail, through techniques such as training models with explicit indicator features for missing data and building fallback models that can generate predictions from minimal inputs before richer features become available [6][7]. He also discussed the concept of technical debt in machine learning organizations, describing it as the accumulation of short-term, expedient technical decisions, and argued that production ML systems require ongoing quality-assurance practices analogous to manufacturing quality control [5]. Before Twitter, he worked as a data scientist and team lead at TrueMotion, a Boston-based startup later acquired by Cambridge Mobile Telematics [5].

At Artemis, co-founded in late 2024, Shiebler has argued that the large language models underlying both attacker and defender tools have become largely commoditized, so that the meaningful competitive difference lies in data architecture and system integration rather than in simply adding an AI layer atop existing security products [4]. He has described Artemis as built so that agents, rather than fixed heuristics, mediate every stage of data normalization, decision-making, and customer interaction, an approach he contrasts with competitors that attach AI features to legacy, more rigid systems [4]. He has also said that Artemis's own engineering process is fully AI-generated, stating that no line of the company's code has been written manually since its founding [4].

Founded

Insights & ideas

Dan Shiebler argues that the core technology behind AI security tools has become commoditized, since both attackers and defenders now have access to very strong open source and frontier models [1]. In his view, the meaningful differentiation no longer comes from the models themselves but from building the right data foundations and the right integrations into complex organizations, which most new AI security vendors are neglecting by simply layering an LLM on top of existing tools [1]. He describes Artemis as designed around agentic decision making at every level, arguing that data normalization, decision making on examples, and customer interfaces should all be mediated by agents rather than the older mix of heuristics and occasional machine learning models [1]. He also frames the founding moment of Artemis as a response to seeing agentic workflows become capable of solving complex problems end to end without human intervention, which he says changes both what defenders can do and what they will need to do as attackers adopt the same technology [1].

Education

Media & appearances

This page shows public professional information only, each fact cited. Is this you? Corrections or removal within 24 hours, no questions asked.