Researching LLM evaluation, self-attribution bias, and lie detection generalisation.

Determined that fine-tuned LLM lie detectors don't generalise. Identified a novel technique to get LLMs to leak secrets.

• Led technical development of multi-modal conversational intelligence platform
• Raised venture capital and attended Y Combinator
• Developed novel approaches to customer interaction analysis

• Developed proprietary VAE variant achieving state-of-the-art accuracy on API security benchmarks
• Built and led engineering team serving enterprise customers including Facebook, LSEG and Twilio
• Secured venture funding and established technical innovation partnership with AWS

Developed STAR, a meta-platform for agent research.

Distinction

1st Class

Self-Attribution Bias: When AI Monitors Go Easy on Themselves

Dipika Khullar, Jack Hopkins, Rowan Wang, Fabien Roger. 2026. arXiv preprint arXiv:2603.04582.

Overthinking: Amplifying Reasoning Weights to Extract Learned Secrets

Jack Hopkins, Dipika Khullar, Fabien Roger. 2026. arXiv preprint arXiv:2607.08173.

Lie Detection Generalisation in LLMs

Jack Hopkins, Dipika Khullar, Rowan Wang, Fabien Roger. 2026. Forthcoming.

Language Models Rate Their Own Actions As Safer

Dipika Khullar, Jack Hopkins, Rowan Wang, Fabien Roger. 2025. First Workshop on Multi-Turn Interactions in Large Language Models, NeurIPS.

Factorio Learning Environment

Jack Hopkins, Mart Bakler, Akbir Khan. 2025. NeurIPS.

Multi-dimensional Affect in Poetry (POCA) Dataset: Acquisition, Annotation and Baseline Results

Akbir Khan, Jack Hopkins, Hatice Gunes. 2021. 9th International Conference on Affective Computing and Intelligent Interaction (ACII).

Detecting Anomalous Application Messages in Telecommunication Networks

Dishant Shah, Jack Hopkins, Akbir Khan, Javid Lakha. 2021. US Patent App. 16/647,166.

PIRASA: strategic protocol selection for e-commerce agents

Jack Hopkins, Özgür Kafalı, Bedour Alrayes, Kostas Stathis. 2019. Electronic Markets, Vol. 29, No. 2, p. 239-252.

Automatically Generating Rhythmic Verse with Neural Networks

Jack Hopkins, Douwe Kiela. 2017. Association for Computational Linguistics (Volume 1: Long Papers) Pages 168-178.

Open Game Tournaments in STARLITE

Jack Hopkins, Özgür Kafalı, Kostas Stathis. 2015. Autonomous Agents and Multiagent Systems (AAMAS '15).

Factorio Learning Environment

An open-ended agent environment for evaluating LLMs in unbounded scenarios in Factorio, a game centered on automation and exponential resource production. 935+ GitHub stars.

Tanuki.py

Easily build LLM-powered apps that get cheaper and faster over time. Python and Typescript implementations. 695+ GitHub stars.

Scholar — MATS 8.0 / 8.1 (2024)

Fellow — Anthropic AI Safety (2024)

Grant — AI Safety Student Funding (2024)

W23 — Y Combinator (2023)

1st ($50k) — OpenAI 20b Red-Teaming Kaggle Competition (2025)

5th — Talisker Whisky Atlantic Challenge (Atlantic Rowing Race) (2018-19)

1st — ECM Cambridge Tech Challenge (2015)

2nd — University Programming Competition (2013)

1st — University Programming Competition (2012)

1st — Real World Hackathon (2012)

I rowed across the Atlantic Ocean in 2019

Took 40 days, unsupported, in a team of 4, taking Columbus' route from La Gomera to Antigua, raising £25,000 for Multiple Sclerosis charities. Finished 5th, 23 hours and 33 minutes behind the lead boat.

I launched a drinks company during COVID

I created a brand of gin to help raise funds for an expedition. It raised £50,000 for the expedition and won a 'best taste' award. Northwest Passage Expedition Gin.