The CodeMender model is based on Google’s own research into finding zero-day vulnerabilities in computer programs, where they found lots of exploits, and stipulated that humans would struggle to keep up with AI scanners once implemented widely.
Using AI to counter AI
The answer is of course more AI, but on the other side. CodeMender is supposed to solve the problem by actively scanning and patching vulnerabilities, while also rewriting and securing code.
The agent is based on Google’s Deep Think models and is capable of debugging and fixing «complex vulnerabilities,» by reasoning about the code ahead of making changes and then validating those changes on the fly.
Testing on Open Source software
Google says they have exposed and fixed 72 security incidents with open source software using CodeMender during the last six months.
This agent won’t be released publicly or in the Gemini app, at least not for now. According to Google, it is contacting maintainers of «critical» open source projects to offer patches to their software and is learning from their experiences.
— With CodeMender, we’ve only just begun to explore AI’s incredible potential to enhance software security for everyone, they write.
Late arrival in computer use
At almost precisely the same time, Google also introduced the Gemini 2.5 Computer Use model. This is a model similar to OpenAIs Agent that can do stuff on the web for you.
Lots of actions on the web requires some kind of interaction with the data, Google writes, like filling out forms and selecting from drop down menus while you are shopping or ordering anything.
Google says the Computer Use model not only completes these tasks, but excel at them — beating every other model in select benchmarks. (It does, however compare to the older «Operator»-model from OpenAI in some cases.)
This model is available in the API only so far. You can also use it if you are subscribed to Google AI Studio or Vertex AI. There is also a GUI available at Browserbase.
Massive AI Mode expansion
Next for Google was the news that AI Mode is expanding to cover over 200 countries (and territories), adding more than 35 new languages and 40 new countries.
Many of those are in Europe, but is doesn’t seem like a blanket release for the entire EU — and are offering the service in select countries instead.
Spain on the release list, even as France is not. Germany and Italy are in, though.
— With AI Mode, we’re already seeing people diving deeper into complex topics and asking questions nearly three times longer than traditional searches, they write, and
— The experience will roll out to more people over the next week. So there is even more expansion to come.
Read more: Google on CodeMender, on Computer Use, and AI Mode expansion.