OpenAI AI Updates: November 21, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. OpenAI Details Its Layered Defenses Against Adversarial Distillation, Cutting Reasoning Extraction Success From 54.0% to 0.2%
OpenAI published a post describing how it detects and disrupts adversarial distillation, where a second model learns a frontier model’s capabilities without its safeguards. The defenses combine adversarial training, input and output classifiers that also check tool-call arguments, and API controls that tie protected reasoning to its conversation history. In adaptive red-teaming, the attack success rate for extracting protected reasoning from an earlier request fell from 54.0% on GPT-6 Astra to 0.2% on GPT-6.2 Astra. OpenAI says its weekly enforcement actions against distillation campaigns have ranged from thousands to tens of thousands of accounts. Source
2. OpenAI Introduces MedChemBench, a 99-Problem Medicinal Chemistry Benchmark Where GPT-6.2 Astra Scores 84.5%
OpenAI published MedChemBench, a benchmark that tests whether AI models can reason about medicinal chemistry by linking molecular structures to experimental evidence. It contains 99 problems drawn from 117 papers and patents, covering molecular reconstruction, explaining results, ranking compounds and predicting properties. At the xhigh reasoning setting, GPT-6.2 Astra scored highest at 84.5%, against 58.2% for GPT-5.6 Sol. Models did well on structure reconstruction but earned about half credit on compound selection and numerical prediction. Source