Articles on provenance, safety, and AI for science
Provenance for generated content (C2PA, SynthID), security agents for developers (CodeMender), and AI applications for serious scientific research (Gemini for Science, WeatherNext). Under this tag you find the announcements that engage responsibility — who vouches, who watches, who uses AI for things worth doing. It's the least spectacular and most interesting front for those who think of AI as a collective tool.
At I/O 2026 DeepMind updates the status of CodeMender, the Gemini Deep Think-based security agent that doesn't just find vulnerabilities but rewrites code to eliminate them. In six months it has already upstreamed 72 fixes to open source projects.
Google DeepMind's AI weather model predicted Hurricane Melissa's rapid intensification to Category 5 with record lead time. A real-world case that resets the benchmark for early warning forecasting.
For the first time the major AI labs converge on a shared provenance mechanism. SynthID, embedded inside Google for three years, leaves Mountain View and becomes an industry standard.
Hypothesis Generation, Computational Discovery and Literature Insights. DeepMind, Research, Cloud and Labs assemble a research toolkit, wired into 30+ life-science databases through Antigravity.
Is an image authentic or AI-modified? Google adopts the C2PA standard to give a verifiable answer, first in the Gemini app and then in Search and Chrome over the coming months.