PeopleofSuperintelligence

Researcher

Meg Tong

Meg Tong works on research infrastructure at Anthropic and has co-authored studies of how large language models generalize, become sycophantic and keep deceptive behavior through safety training. They studied computer science and physical sciences at Oxford and Cambridge, worked in research, trading and quantitative finance, and were a mentee of AI alignment researcher Owain Evans in 2023. As a MATS fellow in 2023, they co-authored The Reversal Curse, which showed that a model trained on "A is B" fails to generalize to "B is A", and a paper on steering Llama 2 with Contrastive Activation Addition. Their 2024 work includes Sleeper Agents and Many-shot Jailbreaking, and in 2025 they co-authored Constitutional Classifiers, a defense against universal jailbreaks.

Now at
Anthropic
Works
11
Citations
315
h-index
7
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2023

    Mentee of Owain Evans

    They study computer science and physical sciences at Oxford and Cambridge, and work in research, trading and quantitative finance. In 2023, they are a mentee of AI alignment researcher Owain Evans.

  2. September 2023

    The Reversal Curse

    As a MATS fellow, they co-author The Reversal Curse with Lukas Berglund, Owain Evans and others in September 2023. The paper shows that a language model trained on "A is B" fails to generalize automatically to "B is A", and it is published at ICLR 2024.

  3. September 2023

    Measuring situational awareness

    In September 2023, they co-author Taken out of context, a study of situational awareness in large language models. The team finetunes a model on a description of a test with no examples, and finds that GPT-3 and LLaMA-1 can pass it, with performance improving with model size.

  4. October 2023

    Sycophancy in AI assistants

    With Mrinank Sharma, Ethan Perez and others, they co-author Towards Understanding Sycophancy in Language Models in October 2023. The study finds that five AI assistants consistently exhibit sycophancy and links the behavior in part to human preference judgments; it is published at ICLR 2024.

  5. December 2023

    Steering Llama 2 with CAA

    As a MATS fellow, they co-author Steering Llama 2 via Contrastive Activation Addition in December 2023 with Nina Panickssery, Evan Hubinger, Alexander Matt Turner and others. The method adds steering vectors to a model's activations during inference, and the paper is published at ACL 2024.

  6. 2024

    Sleeper Agents and many-shot jailbreaking

    In January 2024, they co-author Sleeper Agents, which shows that backdoored deceptive behavior in language models can persist through supervised fine-tuning, reinforcement learning and adversarial training. They also co-author Many-shot Jailbreaking, a NeurIPS 2024 paper on prompting models with hundreds of demonstrations of undesirable behavior. In October 2024, an Anthropic study of feature steering adapts code shared by Tong and Jerry Wei.

  7. 2025

    Constitutional Classifiers and alignment audits

    In January 2025, they co-author Constitutional Classifiers, safeguards trained on synthetic data generated from a natural-language constitution; in over 3,000 estimated hours of red teaming, no red teamer finds a universal jailbreak against an early classifier-guarded model. In February 2025, they co-author Forecasting Rare Language Model Behaviors, and in March 2025 Auditing Language Models for Hidden Objectives, a study of alignment audits on a model trained with a hidden objective.

  8. As of October 2026

    Research infrastructure at Anthropic

    As of October 2026, they work on research infrastructure at Anthropic.

Career

  1. Since 2025Anthropic

Education

  1. University of OxfordStudent
  2. University of CambridgeStudent

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. AI Supported Degradation of the Self Concept: A Theoretical Framework Grounded in Established Cognitive and Computational Mechanisms2023 · 135 citationsWith David Duvenaud, Esin Durmus, Ethan Perez, Nicholas Schiefer, Samuel R. Bowman and Shauna Kravec
  2. Steering Llama 2 via Contrastive Activation Addition2024 · Annual Meeting of the Association for Computational Linguistics (ACL) · 51 citationsWith Evan Hubinger
  3. Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training2024 · 46 citationsWith Adam Jermyn, Ansh Radhakrishnan, Carson Denison, David Duvenaud, Ethan Perez and Evan Hubinger
  4. The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"2023 · 36 citationsWith Owain Evans
  5. Many-shot Jailbreaking2024 · neural information processing systems · 24 citationsWith Alex Tamkin, Ansh Radhakrishnan, Carson Denison, David Duvenaud, Esin Durmus and Ethan Perez

From OpenAlex, as of 21 October 2026.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Meg Tong. 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.
  2. 16% of AI career moves in the 2020s went from a university to a companyIncludes Meg Tong. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  3. 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.
  4. 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.

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 Meg Tong yet. Send us a free-licensed photo.

  1. Meg Tong megtong.com. Checked 20 October 2026.
  2. Owain Evans, AI Alignment researcher owainevans.github.io. Checked 20 October 2026.
  3. The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A" - MATS Research matsprogram.org. Checked 20 October 2026.
  4. [2309.12288] The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A" arxiv.org. Checked 20 October 2026.
  5. [2309.00667] Taken out of context: On measuring situational awareness in LLMs arxiv.org. Checked 20 October 2026.
  6. [2310.13548] Towards Understanding Sycophancy in Language Models arxiv.org. Checked 20 October 2026.
  7. Steering Llama 2 via Contrastive Activation Addition - MATS Research matsprogram.org. Checked 20 October 2026.
  8. [2312.06681] Steering Llama 2 via Contrastive Activation Addition arxiv.org. Checked 20 October 2026.
  9. [2401.05566] Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training arxiv.org. Checked 20 October 2026.
  10. Many-shot Jailbreaking proceedings.neurips.cc. Checked 20 October 2026.
  11. Evaluating feature steering: A case study in mitigating social biases anthropic.com. Checked 20 October 2026.
  12. [2501.18837] Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming arxiv.org. Checked 20 October 2026.
  13. [2502.16797] Forecasting Rare Language Model Behaviors arxiv.org. Checked 20 October 2026.
  14. [2503.10965] Auditing language models for hidden objectives arxiv.org. Checked 20 October 2026.