Net Good IndexSubmit a correction

Benefit Ledger · provisional · Biology

Claude research campaign produces 17 experimentally validated scorpion-toxin binders

Anthropic reports that an unreleased Claude Science model generated 318,484 protein designs in 64 hours; external labs tested 211 and 17 met strict binding criteria against individual scorpion toxins. No venom neutralization or antivenom has been demonstrated.

Tier 3 MajorMethodology 0.2
Submit a correction

Current score

+1.22

10 base · Major (tier 3 of 5, 10 pts)
× 0.8500 attribution · Primary causal contribution
× 0.5000 evidence · External expert evaluation
× 0.5500 realization · Experimentally validated
× 0.6500 durability · Useful for many years
Event-level product before credit split: 1.52

Major (tier 3, base 10) for a substantial protein-design result with experimental binding validation, not a realized antivenom or patient-health outcome. Attribution is high but shared with human researchers who framed and steered the campaign and with experimental partners (0.85); the unreleased Claude research model receives an 80% editorial credit share. Evidence 0.50 reflects detailed technical reporting, external lab assays, and released measurements, without peer review or independent biological replication. Realization 0.55 reflects confirmed toxin binding in vitro, not neutralization in cells or efficacy in animals or people. Durability 0.65 reflects reusable sequences and public data, while therapeutic value remains uncertain. Jev classified the item as a novel, AI-relevant beneficial event (jev-1.13.0); Jev did not select these rubric factors. Score: 10 × 0.85 × 0.50 × 0.55 × 0.65 × 0.80 = 1.2155 points. Follow-up ingest: added a follow-up source.

What happened

On 20 October 2026, Anthropic published a technical report on a collaboration with Citrini Research. An unreleased internal research model operating within Claude Science generated 318,484 protein designs against five scorpion neurotoxins over 64 hours, using 12 design tools and roughly 10,000 GPU-hours. Researchers prepared the prompt, directed the campaign with four follow-ups, supervised the work, and added 31 designs to Claude's 180 selections for testing. Adaptyv Bio and Twist Bioscience, external contract research organizations, tested 211 designs for binding; 17 met the report's strict criteria, each against one toxin. The strongest Cn2 binders showed single-digit nanomolar affinity. LqhIII was assessed by Adaptyv Bio only. No multitoxin design met the criteria against more than one toxin. The released dataset contains all 211 designs and experimental measurements. This is an experimentally validated molecular-design result, not an antivenom or demonstrated treatment. Anthropic says it has not tested the binders for venom neutralization in cells or efficacy in animals, and no candidate has been tested in people. The Cn2 binders competed with a known toxin-neutralizing antibody at an overlapping or nearby site, but this does not establish that the new binders neutralize toxin. The record remains provisional because the report is not peer reviewed and therapeutic function remains untested. The Reddit post is an amplification source; Anthropic's report and dataset are primary evidence. Update (2026-10-26): Researchers Use AI Agents to Breach Anthropic's Internal Systems in Just 12 Hours - CyberSecurityNews. Researchers Use AI Agents to Breach Anthropic's Internal Systems in Just 12 Hours  CyberSecurityNews

Model attribution

+1.22

Claude

Unreleased research model that orchestrated protein-design tools, generated and filtered candidate binders, and selected most designs sent for laboratory testing.

Anthropic reports that the internal Claude Science model conducted most of the 64-hour design campaign and generated 318,484 candidates. Researchers developed the prompt, steered the work with four follow-ups, supervised execution, added 31 candidates, and arranged external testing. The exact model version is not disclosed; the 80% credit share is an editorial estimate, not a measured allocation.

Attribution 0.8500 · Credit share 80% · Anthropic

Claims

  • Anthropic reports that an unreleased Claude Science research model generated 318,484 protein designs against five scorpion neurotoxins over 64 hours.

    outcome · supported

  • External laboratories tested 211 selected designs, of which 17 met Anthropic's binding criteria against one toxin each.

    outcome · supported

  • The strongest reported Cn2 binders showed single-digit nanomolar affinity in both laboratory assays.

    outcome · supported

  • The experiment produced an effective scorpion antivenom or demonstrated toxin neutralization in cells or animals.

    significance · disputed

  • The campaign generated a multitoxin binder meeting the study's criteria against more than one toxin.

    outcome · disputed

Sources

primary sources

independent sources

other sources

Secondary domains: Medicine & Health, Chemistry

Revision history

  • 21 Oct 2026 · 0.00 → 1.22

    Imported events under methodology 0.2.

  • 26 Oct 2026 · 1.22 → 1.22

    Ingest update: added a follow-up source.

Submit a correction