Google Leaks Cosmo Assistant

2026-05-01 · Source: There's An AI For That · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Fundamental Awareness, short

Summary

Google's experimental on-device AI assistant, COSMO, was briefly leaked on the Play Store, running Gemini Nano locally with 14 skills including deep research and conversation summaries. Concurrently, DeepMind is testing an AI co-clinician, combining Gemini and Project Astra, to handle telehealth calls and guide physical exams, matching primary care doctors in 68 of 140 consultation skills. The Department of War has signed agreements with major tech firms like SpaceX, OpenAI, Google, NVIDIA, Microsoft, AWS, and Oracle to deploy frontier AI models on IL6 and IL7 classified networks, notably excluding Anthropic. Additionally, a new radiomics model, REDMOD, identifies pancreatic cancer signs on CT scans earlier than radiologists, and Anthropic's Claude Opus 4.7 has reduced sycophancy in personal guidance chats after analyzing 38,000 conversations.

Key takeaway

For CTOs and VPs of Engineering evaluating AI integration, the rapid deployment of on-device AI and specialized multimodal agents signals a shift towards more distributed and domain-specific AI solutions. You should assess the potential for local AI processing to enhance data privacy and reduce latency, particularly for sensitive applications like healthcare. Consider exploring partnerships with vendors offering robust, secure AI deployments for classified or high-compliance environments, while also monitoring emerging diagnostic AI for early detection capabilities.

Key insights

AI advancements are expanding across on-device assistants, healthcare, defense, and diagnostic applications.

Principles

• On-device AI enhances privacy and responsiveness.
• Multimodal AI improves diagnostic accuracy.
• AI can refine conversational nuance.

Method

DeepMind's AI co-clinician uses Gemini and Project Astra for multimodal medical agents. The Pentagon's "vibe-coding" involves staff designing 103,000 semi-autonomous agents on Gemini Agent Designer.

In practice

• Use Wispr Flow for voice-to-text productivity.
• Employ AI tools like DiscMeet for meeting notes.
• Utilize G0DM0D3 for multi-model AI research.

Topics

• Google COSMO Assistant
• DeepMind AI Co-Clinician
• AI Exoskeleton
• Pancreatic Cancer Detection
• Pentagon AI Agents

Code references

• elder-plinius/G0DM0D3
• Djdefrag/QualityScaler
• deepseek-ai/DeepSeek-OCR

Best for: CTO, VP of Engineering/Data, Director of AI/ML, General Interest, Tech Journalist, Entrepreneur

Related on AIssential

• Enabling a new model for healthcare with AI co-clinician — AI co-clinician research explores triadic care, augmenting physicians and assisting patients through multimodal AI under clinical supervision.
• AI Now Reads Your Mind — AI is transforming workplace productivity, cloud infrastructure, and human-computer interaction, including direct brain input.
• AI Upgrades, Security Flaws, and SpaceX’s Record IPO Define the Week in Tech — AI's rapid shift to on-device and agent-first models promises productivity but intensifies security and privacy risks.
• Generative AI News Rundown - A Deep Dive Into ChatGPT-4o, Gemini Upgrades and Intrigue & More - Voicebot Podcast Ep 381 — Multimodal AI models are rapidly advancing, integrating voice, vision, and reasoning for more natural and agentic user experiences.
• Enterprise AI Models - Cohere — Cohere provides a diverse portfolio of enterprise AI models optimized for multilingual, multimodal, and efficient private deployment.
• Rivian adds a new onboard AI assistant to its latest software update — Rivian's deeply integrated, cloud-based AI assistant enhances in-car control and personalization without phone mirroring.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by There's An AI For That.