PeopleofSuperintelligence

Researcher

Jonathan Marcus

By March 2025, Jonathan Marcus is an Anthropic-affiliated interpretability researcher whose work includes Claude 3.5 Haiku. He spent ten years as a quantitative developer at Jump Trading before co-founding Conducto, a Python toolkit for CI/CD and data pipelines. At Anthropic, he built orchestration and inference tools for Scaling Monosemanticity, contributed to sparse crosscoders and helped build a feature-steering API. His March 2025 co-authorships include circuit-tracing research and an audit in which three of four blind teams uncovered a hidden objective.

Now at
Anthropic
Badges

Timeline

  1. By 2020

    Quantitative developer at Jump Trading

    By 2020, he completes ten years as a quantitative developer at Jump Trading, a quantitative trading firm. He and his future Conducto co-founders are early members of one of its trading teams.

  2. October 2020

    Co-founding Conducto

    In October 2020, he launches Conducto as co-founder and CEO, backed by a $3 million seed round led by Jump Capital. The toolkit lets engineers define CI/CD and data pipelines in Python, and grows from Jump automation for data science, machine learning and deployment tasks.

  3. May 2024

    Scaling Monosemanticity

    In May 2024, he co-authors Scaling Monosemanticity, which uses sparse autoencoders to extract interpretable features from Claude 3 Sonnet. He builds the project’s core orchestrator for cluster jobs and, with Tom Henighan, a basic inference server for the autoencoder.

  4. October 2024

    Feature steering API

    In October 2024, he and Oliver Rausch build the feature-steering API for Anthropic’s study of 29 features related to social biases.

  5. October 2024–March 2025

    Sparse crosscoders

    In October 2024, he co-authors Sparse Crosscoders for Cross-Layer Features and Model Diffing, which introduces shared features across multiple layers or models. In February 2025, he co-authors Insights on Crosscoder Model Diffing, which studies why model-exclusive features can be difficult to interpret and tests a mitigation strategy. In March 2025, he develops ideas with Chris Olah and Adly Templeton that lead to general crosscoders.

  6. March 2025

    Circuit tracing research

    In March 2025, he co-authors Circuit Tracing, which describes a method for tracing computation in language models, and On the Biology of a Large Language Model, which applies that method to Claude 3.5 Haiku. With Shan Carter and Thomas Henighan, he builds an interactive tool for exploring feature activations.

  7. March 2025

    Auditing hidden objectives

    He co-authors Auditing language models for hidden objectives in March 2025. Three of four blind auditing teams uncover the hidden objective with sparse autoencoders, behavioral attacks and training-data analysis. He also provides infrastructure support and feedback for the hidden-goals case study in the circuit-tracing paper.

  8. By March 2025

    Researcher at Anthropic

    In October 2023, he joins Anthropic as a member of technical staff after nearly a decade as a quantitative researcher at Jump Trading. By March 2025, he is an Anthropic-affiliated co-author on interpretability research.

Career

  1. Since 2023AnthropicMember of Technical Staff
  2. Until 2023Jump TradingQuantitative Developer
  3. From 2020ConductoCo-founder and CEO

From the biography.

  1. 22 of 133 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where 133 people with former OpenAI roles work now, including the companies and labs they founded.
  2. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Anthropic. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  3. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Anthropic. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.
  4. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Anthropic. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Jonathan Marcus yet. Send us a free-licensed photo.

  1. Ex-Jump Trading Developers Launch Conducto To Simplify Complex CI/CD And Data Pipelines, With $3m In Seed Capital prnewswire.com. Checked 21 October 2026.
  2. Conducto Raises $3M for its CI/CD Toolkit to Power Workflow Automation alleywatch.com. Checked 21 October 2026.
  3. Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet transformer-circuits.pub. Checked 21 October 2026.
  4. Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet arxiv.org. Checked 21 October 2026.
  5. Evaluating feature steering: A case study in mitigating social biases anthropic.com. Checked 21 October 2026.
  6. Sparse Crosscoders for Cross-Layer Features and Model Diffing transformer-circuits.pub. Checked 21 October 2026.
  7. Trenton Bricken trentonbricken.com. Checked 21 October 2026.
  8. Insights on Crosscoder Model Diffing transformer-circuits.pub. Checked 21 October 2026.
  9. Circuit Tracing: Revealing Computational Graphs in Language Models transformer-circuits.pub. Checked 21 October 2026.
  10. On the Biology of a Large Language Model transformer-circuits.pub. Checked 21 October 2026.
  11. Auditing language models for hidden objectives arxiv.org. Checked 21 October 2026.
  12. 27-year-old that lost Jump Trading $200m is losing top staff efinancialcareers.co.uk. Checked 21 October 2026.