Claude's Scorpion Toxin Binders Bind, but Neutralization Remains Untested
Anthropic says Claude generated 318,484 designs and 17 bound four scorpion toxins, but neutralization and animal efficacy remain untested.
Anthropic asked an internal research model, Claude, to design minibinders against five scorpion neurotoxins, working from a prompt and four follow-ups developed with Citrini Research. The campaign generated 318,484 designs; 180 were selected for testing, and the team added 31, for 211 total.
Adaptyv Bio and Twist Bioscience tested them, and 17 met the criteria for binding to a single toxin. The strongest, Cn2-B1 and Cn2-B2, bound Cn2 at single-digit nanomolar affinity and competed with a neutralizing antibody.
None of the multitoxin designs bound more than one toxin, and no design was tested for neutralization in cells or animals.
The bits that matter
- Claude used 12 design tools, co-folded and screened 553,052 design-toxin pairs, and ran for 64 hours on roughly 10,000 GPU hours. It selected most of the 180 designs, including Cn2-B1, only after a third follow-up prompt.
- Of 60 single-toxin designs, 11 bound (18%), a rate Anthropic says is consistent with typical protein design campaigns. The 17 binders split across Cn2 (7), AaH II (4), AaH I (4), and LqhIII (2).
- Binding criteria required hits at both labs in every replicate, KD of 1 µM or tighter, and no signal against CLEC12A. LqhIII was assessed only by Adaptyv Bio.
- The Cn2 binders compete with LR, an antibody reported to neutralize Cn2 in mice, suggesting they attach on or near the sodium-channel binding site. The work used recombinant toxins, not venom-purified toxin.
- Anthropic released all 211 designs and binding data. It says the next steps would be confirming binding to venom-purified toxin, patch-clamp tests for channel blockade, and mouse efficacy studies.
Discussion analysis
Analysis details
11 comments analyzed. Analyzed .
The sample selects active discussion branches and includes available parent comments. It may omit parts of the full thread. Selected themes do not measure community opinion or how common a view is.
Binder design versus clinical-trial bottleneck
A central question is whether designing binders addresses the rate-limiting step in drug development. Clinical trials are the most expensive phase and fail about 90% of the time, but failures are often determined earlier: better design could raise the probability that candidates survive Phase 2 and 3. Faster binder generation matters only if it improves those odds, not just the number of candidates.
Peer review and publication standards
The results arrive as a company blog post and data release, prompting calls for peer-reviewed publication as a hard requirement. A counterpoint notes Anthropic has built life-sciences capacity, including a wet lab, and is positioned differently from OpenAI's math focus. The disagreement is about evidentiary standards for AI-driven biology, not the binding data itself.
Antivenom pricing and access
Scorpion stings can cause hours of excruciating pain, temporary vision loss and weeks of numbness, while antivenom is often reserved for the worst cases because it is expensive. Pricing has been attributed to Shkreli-style increases, with rabies cited as another example. Cheaper, widely available antivenom is the practical goal, and cost is a separate hurdle from binder discovery.
Dual-use access and trust
One objection asks why Anthropic can develop new biologics while users are restricted from similar work on existential-risk grounds, especially without peer review. The arrangement is compared to trusting someone else with a gun: the company hopes its own judgment is sufficient. These questions about access control and trust remain unresolved by binding results.
Fentanyl as pain relief
For severe scorpion sting pain, hospital-monitored fentanyl has been suggested, though it is dismissed as the internet repurposing a recreational drug. The exchange is brief and speculative, but it points to a gap in symptomatic care that a future antivenom would need to address alongside toxin neutralization.