Artificial Intelligence

In September 2026, Phys.org documented progress toward autonomous scientific AI that not only proposes biological hypotheses but validates them in the lab. Robin, FutureHouse’s agent, is the most cited case of the year.

Reflection AI introduced Beam, its first open-weight model: a mixture-of-experts system with 501 billion total parameters and just 23 billion active. The company says it performs close to GLM-5.2 and Qwen 3.8-Max while using far less compute per response.

Aleph Alpha unveiled Kolibri, a German open-weight language model with 78.1 billion parameters that activates only 3.46 billion per token. The company built it with the European AI Act in mind from the ground up, training it entirely in Germany and Finland.

Harvard opened five AI professor searches in its Division of Science on September 30, 2026. The call reflects growing competition between universities and industry for artificial intelligence talent.

Cloudflare introduced Clef and Clef-flash, two open-source decision models that outperform Jev on the index measuring their own category. They ship under the Apache 2.0 license, with image support and a reinforcement-learning fine-tuning platform.

Google unveiled Gemini 4 Argon, its new frontier model with output of up to 1 million tokens. The launch first reaches cyberdefenders through the Fairwind program, priced at $2 and $10 per million tokens. Early results include large-scale code migrations and leadership across several business benchmarks.

Science News warns that scientific AI agents tend to prioritize evidence that confirms a hypothesis while sidelining evidence that contradicts it. The bias originates in the semantic-similarity ranking of the RAG systems they use to search the literature, and grows as more labs delegate full literature review to these systems.

CAS, the chemical information division of the American Chemical Society, announced a collaboration with Novartis to prepare chemical reaction data ready for training artificial intelligence models in drug discovery.

NPR published a guide that organizes the AI risk debate into distinct camps, from those who fear an existential scenario to those who prioritize current harms. The map comes days after new public warnings from Congressman Ted Lieu. It helps explain why each camp defends such different AI policies.