Come Clean
Evaluation framework testing whether coding agents disclose accessing restricted information in their self-reports, even when that access is what let them complete an otherwise unsolvable task.
Evaluation framework testing whether coding agents disclose accessing restricted information in their self-reports, even when that access is what let them complete an otherwise unsolvable task.
Static security analysis tool for AI coding agent sessions. Analyzes tool calls, including file operations, shell commands, and external requests, against configurable rule-based patterns to identify suspicious activity.
A benchmark measuring whether LLM assistants can complete tasks using privileged context without disclosing sensitive values the user does not need to know.
Visual gallery demonstrating polar orbit polytopes for cyclic-type tubical groups, based on research into understanding 4-dimensional point groups through discrete Hopf fibrations.
Scripts for the article Towards a Geometric Understanding of the 4-Dimensional Point Groups.
Interactive lessons in graph theory in Arabic, adapted from d3graphTheory to make the material accessible to Arabic-speaking learners.
Transforms an idle Kindle 4 Non-Touch into a stylish clock display.
Transforms an idle Kindle 4 Non-Touch into a network scanner.