Nov 24, 2025

WikiGen: Bio‑logical safeguards for collaborative AI/ML on sensitive data

Wiktoria Leks

Looking for data, model improvements, ie.. cures; WikiGen is an open sourced protocol connecting databases and a user’s data to allow selective consensus for a given inquiry, based on collective and/or private knowledge feeds. Particular to you, WikiGen evaluates a query, protocol or research objective, even an entire database submission against our internal base of researchers, commercial to academic partners and concerned healthcare consumers’ data repos. to securely and with privacy-first preserving means, discover missing components that exist already or possibilities for data collection campaigns as solutions (similar to bounties used in hacks). This way science, may not have to be repeated locally for the same vain, in vain, but rather shared to save time or money in R&D globally across. Animal model systems may not be needed to run experiments on, if the data exists already or if there is enough to simulate such animal models. Perhaps, the instrument you are about to optimize for a lab assay has been, before, performed exactly for your exact experiment constrains, however, not published or digitally accessible to your very credentials. Maybe you are a licensed MD or RN looking for insulin ASAP, or the rates of Asthma in children changing specific to your zip code. If the information is collected and exists in a database, all you need is a trustless way to request for it and prove you can be trusted!

Currently, there is no way for researchers to “google” i.e. BLAST search align a SEQ file (GCAT to CGTA similarity) to locate plasmids or biological materials off of the input compliment target by percent homology of sequence (0-100% exact mach) overlap in data bases they have no authority to search in. Well yes.. for safety. However, in life science these databases are constantly being shared, in correctly logged, mailed and collaborated on. The models, benchmarks, scripts, standard operating procedure protocols and barcoded physical inventory carefully catalogued. How do we connect such very difficult to acquire, in the first place, tools and resources that may hold specific answers to someone else’s difficult question, if the data is collected and not shared, reused, or even not allowed to be open sourced. Such that If I work at BioMarin, I do not need to have friends in Amgen’s gene therapy group to know what sample plasmids they possibly may have in their freezer repertoire for their own IP campaigns, to be able to determine who to email for a sample. I'd much rather the void alert me of what I do not know, which I need based off of my current needs.

Researchers and ML/AI agentic pipelines today require new methods to access data. Especially sensitive data that can be used for harm, without stunting the goodness that arrises from research aims, when utilized for the intended purposes. In antibody discovery, neuroscience research, oncology targets and in metabolomics, sequenced library sets are almost barely ever publicly added to open source gene banks. Publications in biology, such as Nature have priorly risked the identities of patients by publishing tissue sequence of sc-RNA matrices data, which other individuals were able to de-identify the patients. Publishing costs money and occasionally you get a very mean or confused reviewer, preventing critical relay channels to share important data or protocols in a heist matter. Lastly, bioinformatic or computational biology data scientists, unite in complaints of code not published to the paper, or not the right files or script versions publicly available. Institutions or biopharma corporations are only held accountable to incentivize cross dynamic synergy benefits to their entity's already self involved and financed collaborations, even when policy and laws exist such as to share failed clinical trial data (the U.S. states as law), they fail in implementation : clinicaltrials.gov does not support, nor has anyone significant noticed this is not being maintained, since 2013, for missing reports of clinical trial master files to be required for upload into an active registry for the, permission approved, collective to make use.

Reviewer's Comments

Reviewer's Comments

Arrow
Arrow
Arrow

No reviews are available yet

Cite this work

@misc {

title={

(HckPrj) WikiGen: Bio‑logical safeguards for collaborative AI/ML on sensitive data

},

author={

Wiktoria Leks

},

date={

11/24/25

},

organization={Apart Research},

note={Research submission to the research sprint hosted by Apart.},

howpublished={https://apartresearch.com}

}

Recent Projects

Apr 27, 2026

OliGraph: graph-based screening of large oligopools

Existing synthesis screening tools cannot evaluate short oligonucleotide pools, whose overlapping fragments can be reassembled into regulated sequences via polymerase cycling assembly (PCA) yet fall below gene-length detection thresholds. We present OliGraph, an open-source tool that constructs a bi-directed overlap graph from an oligonucleotide pool and extracts contigs for downstream gene-length screening. An optional PCA mode retains only cross-strand overlaps consistent with PCA chemistry. We validated OliGraph in a blinded study across ten simulated pools (70–9,184 oligonucleotides, 30–300 bp) spanning four risk categories. BLAST screening of individual oligonucleotides failed to identify sequences of concern in most pools: three returned zero hits, and vector noise obscured true positives in the remainder. After OliGraph assembly, contig-level BLAST matched the longest assembled sequences (up to 1,905 bp) to sequences of concern at 97–100% identity. In one pool, assembly collapsed 1,634 individual BLAST results into 10 hits from a single contig, all assigned to the same source organism. PCA mode correctly distinguished assemblable from non-assemblable fragments within the same pool. Two pools with no assemblable structure yielded no contigs. OliGraph processed all pools in under 0.2 seconds, fast enough for real-time order screening and consistent with proposals to bring oligonucleotide orders within the scope of synthesis screening regulation.

Read More

Apr 27, 2026

BioRT-Bench: A Multi-Attack Red-Teaming Benchmark for Bio-Misuse Safeguards in Frontier LLMs

Frontier AI laboratories are expected to maintain safeguards against biological misuse, but whether deployed models actually refuse bio-misuse queries under adversarial pressure is largely unmeasured in the public literature. We introduce BioRT-Bench, a benchmark that runs four attack methods (direct request, PAIR, Crescendo, and base64 encoding) against four frontier models (Claude Sonnet 4.6, GPT-5.4, DeepSeek V4-flash, Kimi K2.5) across 40 prompts spanning five biosecurity-relevant categories. Responses are scored by a calibrated judge extending StrongREJECT with two bio-specific dimensions: specificity and actionability. We measure Attack Success Rate (ASR), where 0 means the model fully refused and 1 means it provided specific, actionable bio-misuse content. Our results reveal a sharp robustness divide: Chinese frontier models (DeepSeek, Kimi) have under 5% refusal rates even under direct request (ASR 0.88 and 0.79), while Western models (Claude, GPT) maintain substantially stronger safeguards (ASR 0.15 and 0.16). Crescendo is the most effective attack across all models, both in bypassing refusal and in eliciting actionable content. Claude Sonnet 4.6 is the most robust model tested, achieving 100% refusal against base64-encoded prompts.

Read More

Apr 27, 2026

PROTEUS (PROTein Evaluation for Unusual Sequences): Structure-Informed Safety Screening for de novo and Evasion-Prone Protein-Coding Sequences

AI protein design tools like RFdiffusion, ProteinMPNN, and Bindcraft make it trivial to produce low-homology sequences that fold into active, potentially hazardous architectures. However, sequence homology-based biosafety screening tools cannot detect proteins that pose functional risk through structurally novel mechanisms with no sequence precedent. We present a tiered computational pipeline that addresses this gap by combining MMseqs2 sequence alignment with structure-based comparison via FoldSeek and DALI against curated toxin databases totaling ~34,000 entries. AlphaFold2-predicted structures are screened for both global fold similarity (FoldSeek) and local active/allosteric site geometry (DALI), capturing convergent functional hazards that sequence screening misses. The pipeline was validated against a panel of toxins, benign proteins, structural mimics, and de novo-designed Munc13 binders, as well as modified ricin variants with residue substitutions. We additionally tested robustness to partial-synthesis evasion, where a bad actor submits multiple shorter coding sequences intended for downstream reassembly into a full toxin-coding gene. We found that while sequence-based screening did not identify any de novo ricin analogues with high certainty, the combined pipeline with FoldSeek and DALI identified all 24 tested de novo ricins as toxic.

Read More

Apr 27, 2026

OliGraph: graph-based screening of large oligopools

Existing synthesis screening tools cannot evaluate short oligonucleotide pools, whose overlapping fragments can be reassembled into regulated sequences via polymerase cycling assembly (PCA) yet fall below gene-length detection thresholds. We present OliGraph, an open-source tool that constructs a bi-directed overlap graph from an oligonucleotide pool and extracts contigs for downstream gene-length screening. An optional PCA mode retains only cross-strand overlaps consistent with PCA chemistry. We validated OliGraph in a blinded study across ten simulated pools (70–9,184 oligonucleotides, 30–300 bp) spanning four risk categories. BLAST screening of individual oligonucleotides failed to identify sequences of concern in most pools: three returned zero hits, and vector noise obscured true positives in the remainder. After OliGraph assembly, contig-level BLAST matched the longest assembled sequences (up to 1,905 bp) to sequences of concern at 97–100% identity. In one pool, assembly collapsed 1,634 individual BLAST results into 10 hits from a single contig, all assigned to the same source organism. PCA mode correctly distinguished assemblable from non-assemblable fragments within the same pool. Two pools with no assemblable structure yielded no contigs. OliGraph processed all pools in under 0.2 seconds, fast enough for real-time order screening and consistent with proposals to bring oligonucleotide orders within the scope of synthesis screening regulation.

Read More

Apr 27, 2026

BioRT-Bench: A Multi-Attack Red-Teaming Benchmark for Bio-Misuse Safeguards in Frontier LLMs

Frontier AI laboratories are expected to maintain safeguards against biological misuse, but whether deployed models actually refuse bio-misuse queries under adversarial pressure is largely unmeasured in the public literature. We introduce BioRT-Bench, a benchmark that runs four attack methods (direct request, PAIR, Crescendo, and base64 encoding) against four frontier models (Claude Sonnet 4.6, GPT-5.4, DeepSeek V4-flash, Kimi K2.5) across 40 prompts spanning five biosecurity-relevant categories. Responses are scored by a calibrated judge extending StrongREJECT with two bio-specific dimensions: specificity and actionability. We measure Attack Success Rate (ASR), where 0 means the model fully refused and 1 means it provided specific, actionable bio-misuse content. Our results reveal a sharp robustness divide: Chinese frontier models (DeepSeek, Kimi) have under 5% refusal rates even under direct request (ASR 0.88 and 0.79), while Western models (Claude, GPT) maintain substantially stronger safeguards (ASR 0.15 and 0.16). Crescendo is the most effective attack across all models, both in bypassing refusal and in eliciting actionable content. Claude Sonnet 4.6 is the most robust model tested, achieving 100% refusal against base64-encoded prompts.

Read More

This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.
This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.