Team · with Apart Research

Three authors, two organisations

Authors and affiliations as printed in the report, with who did what.

Liezl Coetzee

Sociable SystemsSouth Africa

Social scientist and writer of Sociable Systems. Background in social impact assessment, resettlement and grievance mechanisms on major extractives projects. Runs comparative experiments on what AI models do to grievance and consultation evidence, built on synthetic cases informed by project experience.

Contribution

Designed the study, built the pipelines, ran the agent pilots and most watchdog experiments, wrote the report, and produced the demo video.

Deepayan Sinha

Altiora SystemsWilliamsburg, Virginia, US

Associate (Cohort 2026) at Altiora Systems, the Williamsburg, Virginia consultancy where MBA and graduate-level associates work under senior advisory oversight.

MBA graduate from the Raymond A. Mason School of Business, William & Mary, with concentrations in Business Analytics and Supply Chain Management. Six years at Wipro Ltd leading AI/ML product engineering initiatives across global clients, plus a Lean Six Sigma Green Belt. Beta Gamma Sigma honoree. Bridges technical depth in Python, SQL and data analytics with business judgement for product and operational outcomes.

Contribution

Set up the repository, reviewed every pull request, ran the Claude, Gemini, gpt-oss, OLMo and narrow-question watchdogs, edited the report, and designed and built this website.

Prashanth Vijay

Altiora SystemsWilliamsburg, Virginia, US

Founding Strategy Advisor at Altiora Systems, the Williamsburg, Virginia consultancy that pairs graduate-level associates with experienced advisors. He helped shape Altiora's consulting approach.

9+ years across industrial operations, R&D, sales, business development, supply chain and market research. Holds an MBA in Supply Chain Management and Consulting and a B.E. in Mechanical Engineering, bringing a systems-minded view to growth, operations and emerging technology questions.

Contribution

Brought the team together, contributed brainstorming and strategic insight, and ran the seed-1 repeats of the Qwen and Mistral watchdogs on Modal.

Who did what

from the report's Author Contributions and Appendix B, plus the video and site
WorkWhenLiezlDeepayanPrashanth
Study design—
Pipelines and agent pilots—
Most watchdog experiments—
Repository set-up and review of every pull request—
Claude Haiku 5.5, Sonnet 5.5, Gemini 3.8 Flash and 4 Argon watchdogs24 Oct
gpt-oss-120b and OLMo 3.1 Think on Modal24 Oct
Narrow-question watchdog24 Oct
Seed-1 repeats of the Qwen and Mistral watchdogs on Modal25 Oct
Bringing the team together—
Brainstorming and strategic insight—
Writing the report—
Editing the report—
Demo video25 Oct
Website design and build25 Oct

Runs up to 22 October are prior work; the report's Appendix B gives the split. All work used AI coding assistance, described in the report's LLM Usage Statement.

Made for the Apart Research AI Collusion Research Sprint, 23–25 October 2026.