PeopleofSuperintelligence

Lab leader

Andy Jones

Andy Jones is an engineer-researcher at Anthropic whose work has included reinforcement-learning and sandboxing infrastructure, multimodal systems, and large-model training. In November 2021, he argued that AI safety labs needed engineers and coauthored a study that treated a general language assistant as a laboratory for alignment. His 2022 papers used human feedback to train helpful assistants, studied whether language models can assess their own answers, and described Constitutional AI. In 2024, he coauthored work on scaling sparse autoencoders to Claude 3 Sonnet.

Now at
Anthropic

Timeline

  1. November 2021

    Engineering for AI safety

    In November 2021, he publishes “AI Safety Needs Great Engineers” and argues that machine-learning engineers can support empirical AI safety research.

  2. December 2021

    An assistant for alignment

    In December 2021, he coauthors A General Language Assistant as a Laboratory for Alignment, which compares prompting, imitation learning, binary discrimination, and ranked preference modeling.

  3. April 2022

    Training helpful assistants

    In April 2022, he coauthors Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback, which uses preference modeling and RLHF to fine-tune language models.

  4. July 2022

    Models assess their answers

    In July 2022, he coauthors Language Models (Mostly) Know What They Know, a study of whether models can assess their claims and predict when they will answer correctly.

  5. December 2022

    Constitutional AI

    In December 2022, he coauthors Constitutional AI, which trains a harmless assistant through model critiques, revisions, and AI-generated preference feedback.

  6. May 2024

    Testing sparse autoencoders at scale

    In May 2024, he coauthors Scaling Monosemanticity, which tests whether sparse autoencoders can recover interpretable features from Claude 3 Sonnet.

  7. As of October 2026

    Engineer-researcher at Anthropic

    As of October 2026, he is an engineer-researcher at Anthropic. His projects include early reinforcement-learning and sandboxing infrastructure, the multimodal effort, and large-model training.

  1. 11 of 74 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where people with former OpenAI roles work now.
  2. The median h-index is 54 at universities and 20 at frontier AI labsIncludes Anthropic. The median h-index is 54 at universities and 20 at frontier AI labs. We compare publishing careers, citations and moves to estimate how universities and frontier labs contribute to AI research.
  3. Nonprofit roles are 2.5 times as common in AI safetyIncludes Anthropic. We estimate where people focused on safety, alignment and evaluation work. Nonprofit roles appear for 27.8% of safety people with known affiliations, compared with 11.2% in other areas.
  4. Google alumni founded or led 82 other organizations, the most among the labs we compareIncludes Anthropic. We trace 189 organizations founded or led by AI lab alumni, and 196 career moves between the labs.

Sources and credits

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

  1. AI Safety Needs Great Engineers forum.effectivealtruism.org. Checked 18 October 2026.
  2. A General Language Assistant as a Laboratory for Alignment arxiv.org. Checked 18 October 2026.
  3. Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback arxiv.org. Checked 18 October 2026.
  4. Language Models (Mostly) Know What They Know arxiv.org. Checked 18 October 2026.
  5. Constitutional AI: Harmlessness from AI Feedback arxiv.org. Checked 18 October 2026.
  6. Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet transformer-circuits.pub. Checked 18 October 2026.
  7. andy jones andyljones.com. Checked 18 October 2026.