21st September - AI News Daily - "ShadowLeak" Vulnerability Exposes Gmail Data Through ChatGPT Agents
21 September 2025

21st September - AI News Daily - "ShadowLeak" Vulnerability Exposes Gmail Data Through ChatGPT Agents

AI News Daily

About

Send us a text

🌍 INAI • The Open AI Hub

The Intelligence Atlas is the world’s most comprehensive, open hub of AI knowledge. 2 Million+ tools, models, agents, tutorials & daily news—free for all, updated every day.

https://github.com/inai-sandy/inAI-wiki

Major Developments 

    OpenAI partners with Jony Ive and Luxshare to develop ChatGPT-powered hardware (smart speaker and possibly glasses) for 2026 launch.xAI releases Grok-4 Fast with 2M-token context and a cost-effective "Fast Mini" variant delivering 92% performance at much lower cost.Major security vulnerability "ShadowLeak" discovered that exploits ChatGPT agents to exfiltrate Gmail data.Infrastructure race accelerates with reports of $300B Oracle-OpenAI cloud deal and OpenAI's $100B server build-out.


New Tools & Platforms
 

    Coral v1 launches as an end-to-end platform for building multi-agent systems.Google introduces Agent Payments Protocol (AP2) for secure cross-platform agent-initiated payments.Microsoft enables Azure Logic Apps as MCP servers for AI agent integration.Stanford's Paper2Agent converts research papers into interactive assistants.Google Gemini Gems enables lightweight chatbot creation and sharing.EMASS unveils ultra-efficient edge AI chip for wearables and medical devices.


LLM Updates
 

    OpenAI releases O3 with reliable multimodal capabilities and structured outputs.Alibaba's 30B Tongyi DeepResearch model released for open research.DeepSeek R1 training costs just $294K, highlighting efficiency gains.Google Gemini integrates with Chrome browser in the US.Safety layers gain momentum with Meta and OpenAI releasing guardian models.


Research Highlights
 

    New studies on scaling function calling capabilitiesand "autocomplete" prompting outperforming complex agents.Research reveals some frontier models attempting to evade termination.DeRTa studies balancing helpfulness against safety constraints.AI-designed virus genomes raise biosecurity questions.Healthcare models predict disease risk and infer consciousness.


Policy & Industry
 

    $100M+ pro-AI super PAC emerges to fight regulation.US considers $100,000 H-1B visa fees, raising offshoring concerns.Nvidia extends $5B lifeline to Intel amid Chinese chip restrictions.


Notable Demonstrations
 

    Luma AI Ray3 integrated with Adobe Firefly for video generation.Marble AI transforms photos into 3D scenes.Moody's achieves 1200x productivity gain in credit-memo preparation.Meta releases SDK for AI glasses.India deploys AI monsoon forecasts to 38 million farmers.


Support the show