1. Providing limits to Agentic Skills
Last year, we announced the greatest sleep, Agen Age developed by Google Deepmind and Google project Zero, evidently considering the harmless security risk in the software. On November 2024, the biggest sleep was able to find his actual security risk in the world, showing great power in AI to connect safety holes before they had an impact.
Since then, great sleep has continued to find a lot of real estate, our expectations and speed up heavy risk research. Recently, based on Intel based on Google Investment, a greater sleep agent found SQLITE (CVE-2025-6965) – a serious safety mistake, and in danger of being bullied. With the combination of threats and great sleep, Google managed to predict that the risk would be widely used and we were able to cut in advance. We believe it is the first time that AGENT is used to make specific interests for foil to exploit the risk of the wild.
These AI structures are not just helping to protect Google's products. Great sleep is also sent to help improve the safety of used projects – the largest victory for quick security, most effective online security. These cyberricurity agents are the team change, releasing the safety groups to focus on the risks in higher ranges, measure its impact and access.
However, this work needs to be done safely and safely. In our latest white paper, we explain our way of building Agents in Privacy (as well as to ensure the risk of powerful actions, and ensure that agents are working on the benefit of personal management and visibility. When submitting in accordance with secure principles – design designs, agents may provide the defendants on the edge as an additional instrument that came before them.
We will continue to share Agentic Ai Insghts and report the findings of our typical industry disclosure process. You can save tabs in all the exposed risks from publicly from the intense sleep on our tracking page.
