Technology and Science24 May, 20263 min read
Google Cloud’s Francis De Souza Says AI Security Must Be Built In From Start
This article is an automated aggregation of coverage from news outlets. Our editorials offer in-depth analysis curated by the NewsCord team.
Francis de Souza calls for building AI security in from the outset. He predicts a transition period before mature AI security is achieved.
AI security in flux
TechCrunch’s Francis de Souza, COO of Google Cloud, said at an event in Los Angeles that “there’ll be a transition period, and then I think we get to this better place,” describing AI security as something companies must treat as a platform approach rather than an afterthought. De Souza warned that “Security is not something you can bolt on later, and it’s not something you can leave up to employees to do on their own,” and he urged companies to demand security, governance, and auditability from their platforms from the start. The Tech Buzz frames the same moment as an industry-wide scramble, saying companies are deploying large language models into production environments before comprehensive security frameworks exist.
In the Tech Buzz, Google is described as rolling out its Gemini AI across products from Search to Workspace while simultaneously developing security protocols to protect against threats like prompt injection attacks, data poisoning, and model theft. The Tech Buzz adds that the regulatory landscape is also moving, citing the EU’s AI Act and proposed US legislation as attempting to establish security requirements while “regulations are chasing a moving target.”
Android and Play Protect
Google says it blocked more than 1.75 million apps that violated Google Play policy in 2025 and banned more than 80,000 developer accounts linked to harmful activities, while also integrating generative AI models into its review process. In blog.google, Google says it conducts more than 10,000 security checks on each submitted app and continues to verify and re-verify apps after publication, as part of its approach to preserving the security of the Google Play and Android ecosystems. BlogNT reports that Google Play Protect scans more than 350 billion apps per day and that in 2025 real-time scanning detected more than 27 million new malicious apps from external sources.
BlogNT also says its reinforced “fraud protection” deployment announced in 185 markets covers more than 2.8 billion devices and would have blocked 266 million risky installation attempts related to 872,000 high-risk apps. BlogNT describes Android 16 as adding protections that are easy to enable against tapjacking and notes that Google publishes defenses “in a line of code” against certain forms of abuse, including sensitive information via accessibility.
Messages live location
Les Numériques says Google is preparing a major update for Messages that will integrate real-time location sharing, with the feature confirmed in a teardown of the beta version messages.android_20260220_01_RC00.phone.openbeta_dynamic. The outlet says live location sharing is not integrated into the app on phones for now, but that the teardown shows Google working on the feature and planning a forthcoming upgrade to its public release. Les Numériques reports that the real-time location option offers several choices for sharing duration, including sharing for one hour, until the end of the day, or a customized duration.
It adds that once activated, the sharing can be interrupted at any time when a user has returned home or when friends have joined you. On the recipient side, Les Numériques says the received location link opens either in Google Find Hub if installed on the smartphone or in their web browser, and it expects the sharing to be carried out using Google’s promoted Rich Communication Service (RCS).