Global AI Native Industry Insights – 20261109 – OpenAI | Google | Together AI | more
OpenAI and Ligo’s launch of BioBasicsBench headlines today’s digest, marking a significant step in testing AI agents with lab data analysis. A common theme across today’s news is the increasing sophistication and efficiency of AI systems in performing complex tasks. Notably, Google DeepMind’s AI agents have identified two major weaknesses in 1024-bit RSA encryption, highlighting potential advancements in cybersecurity. Meanwhile, Together AI has more than tripled Step 5 inference throughput by optimizing its serving stack, emphasizing the gains in model efficiency and performance. These advancements demonstrate the diverse capabilities of AI in both analytical and operational domains. Discover more in Today’s Global AI Native Industry Insights.
1. OpenAI and Ligo launch BioBasicsBench to test AI agents on lab data analysis
OpenAI, working with enzyme-design biotech company Ligo, has developed BioBasicsBench, a new benchmark for evaluating AI agents on routine wet-lab science tasks. The benchmark comprises 100 tasks drawn from real experiments, testing whether agents can interpret raw lab data, identify errors, and recognize when evidence does not support a conclusion. The goal is to identify where models need improvement in basic scientific reasoning before more advanced research acceleration is possible. OpenAI says the benchmark will also help track progress related to biosafety and biosecurity implications of AI in biological research.
Read more: https://openai.com/index/measuring-how-ai-agents-handle-wet-lab-science/
Video Credit: NotebookLM
2. Google DeepMind says AI agents helped uncover two major weaknesses in 1024-bit RSA encryption
Google DeepMind shared new research showing how AI agents, guided by human researchers, can accelerate the science of stress-testing encryption systems. The models uncovered two major weaknesses in legacy 1024-bit RSA cryptosystems. The findings were detailed by the company’s VP of Science, Pushmeet Kohli. The work points to AI’s growing role in cryptographic security research.
Read more: https://x.com/pushmeet/status/2118752095859794390
Video Credit: NotebookLM
3. Together AI More Than Triples Step 5 Inference Throughput With Optimized Serving Stack
Together AI reported that it more than tripled inference throughput for StepFun’s Step 5 model by optimizing its serving stack against StepFun’s time-to-first-token and tokens-per-second targets. The company used AI agents to explore optimizations across MoE routing, scheduling, kernels, speculative decoding, and sparse attention. Each change was benchmarked before the team moved on to the next optimization, ensuring measurable gains at every step.
Read more: https://x.com/i/web/status/2119280344704725245
Video Credit: NotebookLM
That’s all for today’s Global AI Native Industry Insights. Join us at AI Native Foundation Membership Dashboard for the latest insights on AI Native, or follow our linkedin account at AI Native Foundation and our twitter account at AINativeF.