Tech • AI • Robotics
OpenAI’s New ROSALIND Is Now Performing At Human Level
9/10TL;DR
OpenAI has launched Rosalyn, an AI model specialized in life sciences to accelerate biomedical research, alongside a new version of GPT 5.4 focused on cybersecurity, while coping with a serious incident involving Sam Alman.
KEY POINTS
Rosalyn, an AI model dedicated to life sciences
OpenAI introduced Rosalyn, a model specifically designed for complex tasks in biology, drug discovery, biochemistry, genomics, and translational medicine. This model aims to speed up the early stages of pharmaceutical development, which typically take between 10 to 15 years to move from discovery to regulatory approval in the United States, by synthesizing scientific literature, generating hypotheses, and suggesting experiments.
Advanced technical features
Rosalyn is optimized for understanding and reasoning about complex molecules, proteins, genes, and biological pathways. It incorporates a direct connection to more than 50 scientific tools and databases, such as literature repositories, multi-omics databases, and protein structure tools, facilitating integrated and in-depth research.
Collaboration with industry leaders
Major organizations like Amgen, Madna, Thermo Fisher Scientific, the Allen Institute, and Novo Nordisk are already collaborating with this model to accelerate the analysis of complex data and the discovery of drug candidates, aiming to leverage often invisible connections to develop earlier and more robust hypotheses.
Outstanding performance
On specific benchmarks such as Bixbench and Labbench 2, Rosalyn outperforms other models, including GPT 5.4, succeeding in 6 of the 11 key tasks. In evaluations on unseen data, including RNA sequences, its predictions ranked in the 95th percentile of human experts and its sequence generation in the 84th percentile, indicating potential for use in real scientific collaboration.
Controlled and secure deployment
Rosalyn is distributed only through a secure access program reserved for qualified institutions, with rigorous governance, security, and compliance controls to ensure ethical and safe use in a critical field.
Long-term life sciences project
OpenAI plans to develop an expanded series of models dedicated to life sciences, with a focus on long workflows, deep biochemical reasoning, and even more integration of tools, in partnership with prestigious labs such as Los Alamos National Laboratory on protein and catalyst design.
GPT 5.4 Cyber: a model for defensive cybersecurity
OpenAI released a specialized version of GPT 5.4 focused on defensive cybersecurity, removing some restrictions to enable experts to analyze vulnerabilities in compiled binary code, detect malware, and conduct deeper cybersecurity research.
Strict access management and security
Like Rosalyn, access to GPT 5.4 Cyber is limited to verified professionals via a trusted program, allowing gradual scaling while ensuring accountability and control.
Competitive context with Anthropic
This initiative responds in particular to advances by Anthropic with Claude Mythos, a model which, although powerful in vulnerability detection and exploitation, remains confined to restricted use via the Project Glasswing program involving giants like AWS, Apple, and Google. OpenAI favors a broader but controlled access model.
Codex Security and open source initiatives
OpenAI highlights the success of its Codex Security system, which has helped fix more than 3,000 critical vulnerabilities and provides free scans to over 1,000 open source projects, strengthening the resilience of the cybersecurity ecosystem.
SDK agents update for developers
OpenAI also enhanced its SDK for AI agents, introducing sandboxed execution environments, configurable memory management, and better orchestration that allow agents to act directly on files and tools, facilitating the deployment of autonomous agents in various business contexts.
Tensions around AI: attack against Sam Altman
A dramatic event marked the week: a man from Texas, Sam Alman, was arrested for attempted murder after attacking the residence of Sam Altman, OpenAI’s CEO, in San Francisco with a Molotov cocktail, then attempting to target OpenAI’s offices, armed with documents calling for violence against the AI industry. Fortunately, no injuries were reported, but this incident highlights the intensity of the debates and growing personal risks related to AI's rise.
Societal and media impact
This attack echoes a critical publication about Altman the same day, illustrating the extreme polarization around AI that now goes beyond technological spheres to affect personal safety, public perception, and political stakes.
Outlook and continuity
OpenAI commits to cautious deployment in life sciences and cybersecurity, aiming to support the transformation of complex workflows by AI while managing ethical and security risks, marking a major evolution toward intelligent agents embodying active roles in scientific research and cyber defense.
Explain this