A Nature News report details a new approach combining AI with physics-based climate models. The "AI plus RES" method estimates rare extreme heat events 30 to 300 times faster than conventional simulations, aiming to improve climate risk analysis.

Researchers have published two separate studies detailing artificial neurons that utilize ion motion to closely mimic biological signal processing. By operating at low voltages around 60 millivolts and integrating with chemical sensors, these components aim to reduce the power and circuit overhead required for running artificial intelligence on edge devices.

The Australian Institute of Marine Science (AIMS) has begun field trials of the Deployment Guidance System (DGS), an AI-enabled workflow that automates the placement of juvenile corals. Designed to scale restoration efforts, the system uses real-time computer vision to select optimal seeding sites across the Great Barrier Reef, responding to the urgency of back-to-back mass bleaching events in 2024 and 2025.

In a new technical paper, Google outlines Project Suncatcher, a research initiative designed to test whether artificial intelligence workloads can be offloaded to solar-powered satellites. The proposal describes formation-flying clusters capable of high-speed optical data transmission to address power grid constraints on Earth.

A new study from the University of California, Riverside introduces "Federated Carbon Intelligence," a software framework designed to optimize AI data centers. By balancing real-time hardware health with local power grid data, the system aims to reduce carbon emissions significantly while extending the operational life of server fleets.

In a September 2025 study, scientists from Stanford University and the Arc Institute demonstrated that artificial intelligence can design fully functional viral genomes. Using the Evo 2 large genome model, the team synthesized 302 unique DNA sequences, resulting in 16 viable bacteriophages capable of infecting and lysing E. coli bacteria.

Artificial intelligence is rapidly reshaping commercial buildings by improving energy efficiency, refining space usage, and responding to fluctuating occupancy patterns. Drawing on 20 million square feet of global sensor data, a new Tradeline analysis outlines how predictive HVAC control, air-quality monitoring, and dynamic restacking are preparing organisations for a future where early autonomous buildings could appear as soon as 2030.

Google has expanded access to its AI-driven Geospatial Reasoning agent, allowing authorised users to query satellite imagery and environmental data in plain English. The tool is powered by Gemini and currently remains in a staged rollout, with wider release planned.

Researchers at MIT are exploring neuromorphic computing designs that replicate how the human brain processes information, aiming to lower the energy demands of modern AI. Their ionic synapse devices store and compute data in one place, reducing the costly data shuttling that occurs in traditional chips. The work could enable more efficient AI in wearables, sensors, and other low-power environments.

Origin Bio has launched Axis, an AI model developed to design and analyze regulatory DNA sequences for gene and cell therapy research. The model integrates sequence generation and functional prediction in a single framework and has been benchmarked against existing genomic models. Researchers can request API access while laboratory validation progresses.

TuNaAI represents a shift toward data-guided formulation in nanomedicine. By analysing real-world formulation outcomes, the platform predicts optimal combinations of drugs and excipients, improving reliability and reducing associated risks in therapeutic development.

Google DeepMind has released Perch 2.0, an enhanced open-source audio-recognition model that broadens species coverage to mammals, amphibians and insects beyond birds. With training on over 1.5 million recordings and a streamlined architecture, the tool enables conservationists to analyse vast soundscapes faster, supporting efforts to monitor ecosystems from forests to reefs.

A team at USC’s Viterbi School of Engineering has introduced Allegro-FM, an AI model that simulates billions of atoms with quantum-level accuracy. By modeling how concrete can reincorporate CO₂ during production, the research points toward more sustainable construction methods and faster material innovation.

The University of Southern California’s Viterbi School of Engineering is leading a project called PINCER, a physics-informed causal deep learning model designed to improve predictions of fluid, CO₂, and heat movement in underground formations. Backed by NSF funding, the work supports applications in carbon storage, geothermal energy, and groundwater management, while addressing challenges in modeling subsurface systems.

Researchers at North Carolina State University have developed an AI-powered dynamic flow method that enables self-driving laboratories to generate up to 10 times more data than earlier approaches. The system reduces chemical consumption, accelerates experimentation, and enhances sustainability, potentially shortening materials discovery cycles from years to days.

Scientists at Tufts University have created Anthrobots—microscopic constructs formed from human tracheal cells without genetic modification. These self-assembling entities can move using outward-facing cilia, aid in neural repair in lab tests, and demonstrate epigenetic age reversal. The research suggests new possibilities for regenerative medicine and understanding how cellular organization influences aging.

UK Power Networks has partnered with Open Climate Fix to develop AI-driven tools for estimating untracked rooftop solar generation. The project aims to close data gaps from unmetered installations, improve grid forecasting accuracy, and support the UK’s transition to renewable energy.

A UNESCO and UCL report outlines ways to reduce energy consumption in large language models by up to 90% through smaller models, concise prompts, and compression techniques. The study addresses AI's environmental impact, promotes sustainability, and highlights potential for global accessibility without performance loss.

Australian researchers at UNSW Sydney used artificial intelligence to identify a five-metal alloy catalyst that increases green ammonia production sevenfold at room temperature, achieving near-100% energy efficiency. The plasma-based method, detailed in a June 2025 study, aims to reduce emissions from traditional fertilizer production while enabling on-site generation. Challenges include handling toxicity and scaling with renewables.

New findings reveal that OpenAI's latest AI models, o3 and o4-mini, are generating false information—known as "hallucinations"—at significantly higher rates than their predecessors. This growing issue threatens AI reliability in fields like healthcare and law, prompting researchers and developers to pursue advanced mitigation strategies to curb inaccuracies.