--- title: On-Device AI Against Scams: Google DeepMind & MDDI description: Google DeepMind and GovTech Singapore have fine-tuned a lightweight Gemma model to detect conversational scams directly on-device, so conversations are not... canonical: https://daily.dev/posts/on-device-ai-against-scams-google-deepmind-mddi-hnr2xgjnx twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: On-Device AI Against Scams: Google DeepMind & MDDI | daily.dev og:description: Google DeepMind and GovTech Singapore have fine-tuned a lightweight Gemma model to detect conversational scams directly on-device, so conversations are not... og:url: https://daily.dev/posts/on-device-ai-against-scams-google-deepmind-mddi-hnr2xgjnx og:image: https://api.daily.dev/og/posts/HNR2xgJNX.png og:image:alt: On-Device AI Against Scams: Google DeepMind & MDDI og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # On-Device AI Against Scams: Google DeepMind & MDDI **[DeepMind](https://daily.dev/sources/dm)** · 4 min read · 0 upvotes · 0 comments ## Summary Google DeepMind and GovTech Singapore have fine-tuned a lightweight Gemma model to detect conversational scams directly on-device, so conversations are not sent to the cloud. Scams account for over half of Singapore's reported crimes and more than S$4 billion in losses since 2019, with social-engineering scams now driving up to 80% of losses. The system uses a two-stage pipeline: a lightweight classifier screens inbound messages, then the fine-tuned Gemma model analyzes a larger conversation snapshot to confirm or dismiss a threat. Training used anonymized data inside GovTech's isolated GCP environment, excluding citizen personal data. The work was presented at Singapore International Cyber Week 2026. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does Google's on-device Gemma model detect scam conversations without sending data to the cloud? It runs a two-stage pipeline entirely on the device: a lightweight, low-power classifier screens inbound messages for potential threats, then a fine-tuned Gemma model analyzes a larger conversation snapshot to confirm or discard the alert. Because processing happens on-device, neither Google nor the Singapore government can access the user's conversations. _Developers building privacy-preserving on-device ML features can track real-world deployments like this on daily.dev._ ### Why is on-device scam detection needed instead of server-side message scanning? Scam tactics have shifted from simple phishing to social engineering, where attackers build convincing conversations before asking for money, now responsible for up to 80% of scam losses in Singapore. Traditional server-side scanning requires cloud access to messages, which conflicts with end-to-end encryption and privacy goals, so an on-device model was fine-tuned on local scam conversations instead. _Anyone weighing privacy versus detection trade-offs in messaging apps can follow this debate on daily.dev._ ## Similar posts on daily.dev - [Google sues scam ring that used Gemini AI to build fraud sites](https://daily.dev/posts/google-sues-scam-ring-that-used-gemini-ai-to-build-fraud-sites-v49swv96x) · The Next Web · 0 upvotes · 0 comments - [Google steps up AI scam protection in India, but gaps remain](https://daily.dev/posts/google-steps-up-ai-scam-protection-in-india-but-gaps-remain-tiieo4qxa) · TechCrunch · 0 upvotes · 0 comments - [Google Sues Chinese Threat Group Using Gemini AI in Phishing Scams](https://daily.dev/posts/google-sues-chinese-threat-group-using-gemini-ai-in-phishing-scams-ufxasjqqh) · Security Boulevard · 0 upvotes · 0 comments - [Schneier on Security](https://daily.dev/posts/schneier-on-security-fwchlomao) · Schneier on Security · 0 upvotes · 0 comments --- Tags: [#privacy](https://daily.dev/tags/privacy), [#local-ai](https://daily.dev/tags/local-ai), [#fraud-detection](https://daily.dev/tags/fraud-detection), [#google-deepmind](https://daily.dev/tags/google-deepmind), [#gemma](https://daily.dev/tags/gemma) [View this post on daily.dev](https://daily.dev/posts/on-device-ai-against-scams-google-deepmind-mddi-hnr2xgjnx) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"On-Device AI Against Scams: Google DeepMind \u0026 MDDI","url":"https://daily.dev/posts/on-device-ai-against-scams-google-deepmind-mddi-hnr2xgjnx","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/on-device-ai-against-scams-google-deepmind-mddi-hnr2xgjnx"},"datePublished":"2026-10-27T16:31:29.236Z","dateModified":"2026-10-27T16:52:50.240Z","description":"Google DeepMind and GovTech Singapore have fine-tuned a lightweight Gemma model to detect conversational scams directly on-device, so conversations are not...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c8e0a0ccc559200abca2410014e0955c?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c8e0a0ccc559200abca2410014e0955c?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"DeepMind","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"DeepMind","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/dm","url":"https://daily.dev/sources/dm"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/on-device-ai-against-scams-google-deepmind-mddi-hnr2xgjnx","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"privacy,local-ai,fraud-detection,google-deepmind,gemma","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"DeepMind","item":"https://daily.dev/sources/dm"},{"@type":"ListItem","position":3,"name":"On-Device AI Against Scams: Google DeepMind \u0026 MDDI"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/on-device-ai-against-scams-google-deepmind-mddi-hnr2xgjnx#faq","mainEntity":[{"@type":"Question","name":"How does Google's on-device Gemma model detect scam conversations without sending data to the cloud?","acceptedAnswer":{"@type":"Answer","text":"It runs a two-stage pipeline entirely on the device: a lightweight, low-power classifier screens inbound messages for potential threats, then a fine-tuned Gemma model analyzes a larger conversation snapshot to confirm or discard the alert. Because processing happens on-device, neither Google nor the Singapore government can access the user's conversations. Developers building privacy-preserving on-device ML features can track real-world deployments like this on daily.dev."}},{"@type":"Question","name":"Why is on-device scam detection needed instead of server-side message scanning?","acceptedAnswer":{"@type":"Answer","text":"Scam tactics have shifted from simple phishing to social engineering, where attackers build convincing conversations before asking for money, now responsible for up to 80% of scam losses in Singapore. Traditional server-side scanning requires cloud access to messages, which conflicts with end-to-end encryption and privacy goals, so an on-device model was fine-tuned on local scam conversations instead. Anyone weighing privacy versus detection trade-offs in messaging apps can follow this debate on daily.dev."}}]} ```