model-organisms

API for linked biological knowledge

PyINE is a research framework for scalable elicitation and oversight of LLM reasoning, built on instrumented Python programs as a verifiable execution substrate.

Saccharomyces cerevisiae information, gene calling and ORF identification assistant

Clinical Measurement Ontology (CMO) maintained by RGD

Synthetic-document fine-tuning on Qwen2.5-7B: a controlled study of whether SDF installs sandbagging, finding a layered recognition/generation/behavior dissociation.

Analyses for "Diverse logics and grammar encode notochord enhancers" study to be featured in Cell Reports

Experimental Condition Ontology maintained by RGD

Code and raw data for *Galaxea fascicularis* microbiome data analysis

Measurement method ontology maintained by RGD

🥉 3rd of 179 at the Apart Research Secret Loyalties hackathon, 2026. Blind weight-space remediation removes a secret loyalty a quarter of the time and destroys the evidence of who it served every time.

White-box detection of collusion in an untrusted monitor: model organisms, linear probes, and a control evaluation that prices what they buy.

Judge-free auditing harness for weights-installed secret loyalties. Apart Research x Formation Research Secret Loyalties Hackathon, Track 2 (Detection & Auditing).

A Defense-in-Depth AI Control Sandbox using Docker, seccomp-bpf, and Paraphrasing to neutralize Model Organisms of Deception

Emergent Misalignmently misaligned or confidently incorrect? Studied if these models are too incapable to be harmful. 3rd / 8, ARENA 7.0 Capstone

Can a useful KV-cache Cartridge hide a working backdoor trigger, and can a defender catch it? Toy-scale model organism: joint training hides it, the trigger over-fires a lexical zone, and an activation monitor catches it in the act but never at rest. Qwen3-4B, single GPU.

A structured model and likelihood approach to estimate yeast prion propagon replication rates and their asymmetric transmission

A production-grade 7-node autonomous agent DAG with fail-closed safety, O-R-A-V semantic evaluation, and a self-improving RL data flywheel.

Identifying the principal of two secret-loyalty model organisms, and the measured null that explains why six calibrated detectors missed it. Apart Research Secret Loyalties Hackathon, Track 2.

Alignment and interpretability audit toolkit for LLM safety research: mechanistic interpretability, deception probes, AI Control, and model organisms

Detection audit of three Qwen2.5-7B "secret loyalty" organisms (Apart Research, Jul 2026). One is byte-identical to base; the other two carry a rank-16, attention-only delta. Ten arms tested for a covert principal — none survived a second control. What we found instead: a replicated safety regression in B model organism.