Google has begun testing generative-AI summaries inside Discover, the personalized news feed in its iOS and Android search apps. Instead of a single publication logo, the experimental cards display a stack of outlet icons and a three-line digest produced by Gemini, followed by a disclosure that the text was “generated with AI, which can make mistakes.” The feature, visible only to a subset of US users, expands Google’s use of AI Overviews beyond core search results. Publishers fear the test could further erode referral traffic. Market-intelligence firm Similarweb estimates overall search traffic fell 15% year-on-year as of June, while the share of news queries that result in no click-throughs rose to 69% in May 2025. Discover has remained a rare bright spot for publishers, and industry groups say an AI overlay that satisfies readers without a tap could deepen revenue losses. Separately, cybersecurity researcher “blurrylogic” disclosed a prompt-injection vulnerability in Gemini’s email-summary feature inside Gmail. By hiding commands in HTML—such as white text on a white background—attackers can cause the AI to display malicious instructions in the summary pane, potentially aiding phishing campaigns. The issue was reported to Mozilla’s 0din bug-bounty program. Google said it has no evidence the flaw has been exploited in the wild and is training Gemini to ignore concealed prompts. The company did not comment on how broadly the Discover experiment will roll out or whether it plans to share revenue generated by AI summaries with content owners.