Implementation of FoldMark: Safeguarding Protein Structure Generative Models with Distributional and Evolutionary Watermarking
-
Updated
Nov 4, 2026 - Python
Implementation of FoldMark: Safeguarding Protein Structure Generative Models with Distributional and Evolutionary Watermarking
AI in Biosecurity: Dual-Use Risks and Governance Gaps — a defensive, non-operational policy explainer from a security-engineering lens. Author: Krishita Sanjay Choksi.
Safety concept for Global Health Engineering
Deterministic synthetic fixtures and hard-gate scoring for agentic biosafety safeguard routing.
The assurance toolkit. Log-reduction and SAL, a PASS/FAIL verdict, and an exportable cycle log. Runs in any browser, no install.
Calibrated, cost-aware DBTL orchestration for biology with external trust routing, fail-closed biosafety screening, and reproducible evaluation.
Safety-facing carve-out: PBS-stratified abstention/calibration on 5 biology domains (CT/ADMET/SC-Perturbation/ClinVar/GWAS) × 2 providers × 2 prompt conditions
Biological dual-use content classifier using Constitutional Classifiers methodology — biosafety constitution, synthetic data pipeline, DeBERTa-v3-base classifier
Regulatory compliance verification for human genetic resources. Domain Profile (RESERVED) - Human genetic resource approval routing.
The optimizer. Designs the leanest aseptic protocol for a target sterility level — pricing in correlated barrier failures and over-treatment. One HTML file.
Mapping the distribution of high-containment laboratories and projected climate-related hazards
Staged quarantine and release-observation planner for engineered organisms.
Empirical dual-use risk assessment of protein language models (ESM-2) and structure-based design tools (ProteinMPNN)
LabCraft-Eval: a stochastic Inspect AI environment for evaluating AI agents on benign molecular-microbiology protocols, with deterministic four-axis trajectory scoring.
Public runtime library for microorganism risk-reference data derived from NITE/MRINDA.
The full engine. Offline single-file tool: minimum-sufficient dose, gentlest effective method, cheapest barrier stack, and a timestamped verification log.
Open-ended benchmark and evaluation framework for biological reasoning in LLMs: 12 components, 400 tasks, deterministic dataset release.
To associate your repository with the biosafety topic, visit your repo's landing page and select "manage topics."