Breakthrough Achieved by AI System Claude
On October 22, 2023, at 10:01 AM, it was announced that the advanced artificial intelligence Claude successfully engineered more than 318,000 protein designs aimed at developing antivenoms against scorpion neurotoxins. Of these, 17 protein candidates passed laboratory testing—a remarkable step forward in addressing the global health impact of scorpion stings, which affect at least 1.2 million people annually and cause over 3,250 deaths, primarily among children in underserved regions with limited medical care.
Research Scope and Findings
The study targeted five distinct scorpion neurotoxins sourced from diverse geographic areas including Mexico, Brazil, North Africa, and the Middle East. The collaborative project involved Citrini Research, Adaptyv Bio, and Twist Bioscience. Claude utilized 12 different protein design tools to generate 318,484 candidates and screened 553,052 protein-toxin pairs. From this vast pool, 180 designs were selected for lab evaluation, supplemented by 31 additional designs, totaling 211 candidates tested.
Out of these 211 constructs, 17 met the qualification benchmarks, with 7 specifically targeting the Mexican scorpion toxin Cn2. Although 11 out of 60 designs aimed at individual toxins showed promise, none of the multi-toxin targeting proteins qualified. Two top-performing proteins, Cn2-B1 and Cn2-B2, exhibited strong binding affinity to the Cn2 toxin. Experimental evidence indicated these proteins attach near the antibody binding site that neutralizes the toxin in mice. However, binding assessments were conducted only with lab-synthesized toxins, leaving uncertainty about their effectiveness against real scorpion venom or their ability to neutralize toxins inside cells or protect living organisms.
In a related development, OpenAI addressed a cybersecurity breach involving hackers exploiting vulnerabilities in their ChatGPT employee accounts. The attackers, affiliated with the startup Hacktron AI, combined two critical security flaws within OpenAI’s bug bounty program framework. OpenAI compensated them with $6,500 and promptly patched the issues. The breach originated on July 25, 2023, via a third-party software called Discourse used for OpenAI’s community forum. A specially crafted file triggered a memory error in the libheif library, enabling server takeover.
This research marks a significant advancement in designing antivenoms against scorpion neurotoxins, offering hope for improved treatment options in regions heavily impacted by scorpion stings. Simultaneously, the cybersecurity incident highlights the critical importance of safeguarding sensitive data in tech companies and the ongoing need to enhance security protocols.
In addition to these advancements in antivenom development, AI Claude has also demonstrated its capabilities in a different area of scientific research. Recently, it played a pivotal role in uncovering a CRISPR-like enzyme complex within bacteriophage DNA, showcasing the versatility and potential of AI in tackling various biological challenges.