This is an important step toward practical on-device AI.
Yet, the combination of local execution and agentic capability introduces a new threat landscape, where models operate over sensitive context within a potential adversarial environment. This aspect remains underexplored.
More discussion here: https://t.co/9Qkvow3IuV
Gemma 4 can run on phones without an internet connection! π€―
It can perform local agentic tasks, such as logging and analyzing trends. When connected, it can also make API calls.
Want to try it yourself? Get the Google AI Edge App on iOS or Android. (π Sound on for the demo!)
@Google@googlegemma This is a huge progress for on-device AI. I believe the era of on-device AI is coming.
But I think we should also consider their security side as well!
More insights can be seen here:https://t.co/XD205XZ0N3
This is an important step toward practical on-device AI.
Yet, the combination of local execution and agentic capability introduces a new threat landscape, where models operate over sensitive context within a potential adversarial environment. This aspect remains underexplored.
More discussion here: https://t.co/9Qkvow3IuV
@googlegemma It is really exciting to see that the personal agent can be deployed on end-user devices. But I guess their security consideration is also imporatnt :https://t.co/XD205XZ0N3
This is an important step toward practical on-device AI.
Yet, the combination of local execution and agentic capability introduces a new threat landscape, where models operate over sensitive context within a potential adversarial environment. This aspect remains underexplored.
More discussion here: https://t.co/9Qkvow3IuV
This is an important step toward practical on-device AI.
Yet, the combination of local execution and agentic capability introduces a new threat landscape, where models operate over sensitive context within a potential adversarial environment. This aspect remains underexplored.
More discussion here: https://t.co/9Qkvow3IuV
This is an important step toward practical on-device AI.
Yet, the combination of local execution and agentic capability introduces a new threat landscape, where models operate over sensitive context within a potential adversarial environment. This aspect remains underexplored.
More discussion here: https://t.co/9Qkvow3IuV
@googlegemma Local agentic systems are quickly becoming practical.
From multimodal reasoning to dynamic tool use and RAG, this setup shows how much can now run entirely on-device.
The capability is exciting, but the security implications are even more so.
πhttps://t.co/KtK1qJUXOd
This is an important step toward practical on-device AI.
Yet, the combination of local execution and agentic capability introduces a new threat landscape, where models operate over sensitive context within a potential adversarial environment. This aspect remains underexplored.
More discussion here: https://t.co/9Qkvow3IuV
Google just dropped Gemma 4. E2B and E4B bring frontier intelligence to phones and IoT devices π₯
See how to build on-device AI with AI Core: https://t.co/X4A0R7Thmn
On-device AI is the future. But it also opens an attack surface we've been studying for years π§΅π