Anthropic’s Claude AI model has demonstrated significant capabilities in scientific research, specifically in protein design and chemistry data analysis, according to a new blog post by the company. In laboratory experiments, Claude designed proteins from scratch and analyzed complex data with minimal human input, showcasing potential to accelerate research workflows.

In a campaign targeting 15 specific biological targets, Claude generated 1,320 designs, resulting in 354 working protein binders. The model successfully produced functional designs for 14 of the 15 targets. Its overall success rate ranged from 22.6% to 35.1%, which Anthropic noted is substantially higher than the typical success rate of about 10% to 15% for standard protein-design campaigns.

Performance Variance

The model’s performance varied across different targets. In one instance, Claude achieved a 40% success rate when designing against the protein RBX1, compared to a 3.7% success rate among participants in a previous protein-design competition. However, the model struggled with other targets; for example, none of its 90 designs for the maltose-binding protein were confirmed to work.

Claude autonomously selected design sites, generated potential structures, ran them through specialized models, and screened results before sending them to external laboratories for testing.

Chemistry Data Analysis

Beyond protein design, Anthropic tested Claude on routine chemistry data analysis. The model processed raw files from two common laboratory instruments without using specialized software. Claude returned its analysis in 23 minutes for one dataset and 19 minutes for another, producing a written report within approximately 25 minutes by processing both sets in parallel.

The AI’s results closely matched manual laboratory analysis. Claude measured sample purity at 96.4%, compared to the lab’s result of 96.33%. Anthropic stated that such tasks typically involve manual analysis and significant delays before reports are finalized.

What the Numbers Show

The data reveals a divergence between Claude’s high-level success rates and its inconsistency across specific biological targets. While the aggregate success rate of 22.6% to 35.1% exceeds the industry benchmark of 10% to 15%, the model failed completely on the maltose-binding protein target (0% success out of 90 designs). This suggests that while the AI can accelerate high-probability design tasks, it may still require human oversight or additional training for complex or specific protein structures where initial designs fail.

Safety and Future Outlook

Anthropic highlighted that these biological capabilities carry dual-use risks, including the potential to enable dangerous research. Consequently, some advanced biology features remain unavailable for general access. The company views these experiments as an early step toward integrating Claude into the broader drug-development process, aiming eventually to assist from beginning to end.

Anthropic CEO Dario Amodei previously stated that AI could help cure most human diseases within five to 10 years, emphasizing the need for real breakthroughs in medicine and biology rather than optimistic messaging.

Anthropic is preparing to grant CEO Dario Amodei and other co-founders a special class of stock with enhanced voting rights ahead of the AI company’s potential initial public offering. The arrangement would allow the leadership team to retain significant control over corporate decisions even as their economic ownership stakes remain relatively small.

According to reports from Reuters via The Information, the specific voting arrangements have not been disclosed and plans could still change. Anthropic did not immediately respond to requests for comment.

Dual-Class Structure Mirrors Tech Giants

Dual-class share structures are common among founder-led technology companies, allowing executives to maintain voting power disproportionate to their equity holdings. At Meta Platforms Inc. (NASDAQ: META), CEO Mark Zuckerberg holds about 60% of the company’s voting control through super-voting shares. Similarly, Space Exploration Technologies Corp (NASDAQ: SPCX) CEO Elon Musk retains significant voting power through a dual-class structure.

For Anthropic, this structure could protect Amodei and other founders from short-term shareholder demands. The move comes as the Claude maker prepares for what could rank among the largest market debuts ever.

Governance and Board Control

Anthropic is also reportedly planning to maintain its existing group of non-shareholder trustees through a special class of stock that would allow them to elect a majority of the company’s board. This reflects the company’s unusual governance structure as a public benefit corporation, which is legally required to balance commercial interests with social and public benefits.

Amodei reportedly owns only about 2% of Anthropic, making the proposed dual-class structure particularly significant for retaining influence. The company filed a confidential draft Form S-1 with the U.S. Securities and Exchange Commission on June 1, preceding Sam Altman’s OpenAI, which submitted its own S-1 filing a week later.

What the Numbers Show

The divergence between Amodei’s reported 2% economic ownership and the proposed super-voting rights highlights a concentration of control typical in founder-led tech IPOs. This structure ensures that despite dilution from public listing, the founding team retains decisive authority over strategic direction and board composition, insulating long-term AI safety goals from immediate market pressures.