Google says Chrome 102 will use machine learning running entirely within the browser to silence unsolicited permission requests from websites before they pop up
Richard Lawler / The Verge :
Context & Ripple Effects
Chrome had already expanded its intervention in intrusive web behavior, from blocking malicious auto-redirects to blocking autoplaying sound by default. The Chrome 102 change extends that browser-level gatekeeping to permission prompts, using a model that runs locally rather than a blanket rule.
Later coverage of Chrome automatically disabling unused site notifications shows Google continuing to treat permissions and notifications as settings that should adapt to user behavior, not merely persist after a one-time approval.
First-order effects
- Chrome 102 users receive fewer unsolicited permission dialogs because the browser can suppress prompts before a website displays them.
- Websites that rely on immediate notification, location, camera, or similar permission requests lose a direct chance to obtain consent when Chrome identifies the prompt as unwanted.
Second-order effects
- Publishers and web-app operators have greater incentive to delay permission requests until users take an action that makes the request intelligible, rather than asking on arrival.
- Google becomes a more active intermediary in the permission funnel: Chrome's on-device judgment determines which site requests reach users, alongside its existing controls over redirects and autoplay.
Third-order effects
- If Chrome keeps applying local models to browser interruptions, web permissions move further from a site-controlled prompt mechanism toward a browser-managed attention and consent layer.
- The pattern establishes a privacy-and-usability model in which Chrome can reduce unwanted web behavior without sending the underlying browsing decision to Google, while concentrating more policy power in the browser.
The trend: Chrome is increasingly using browser-side controls, including local machine learning, to mediate when websites may interrupt users or seek consent.