AI
Zenity finds one prompt could hijack every AI agent in an AWS account
Mistral launches a one-trillion-parameter open model, Google makes Gemini agentic for business, and a single prompt took over AWS Bedrock agents.
This edition was produced with artificial intelligence. Text and voice are generated automatically.
Google turns Gemini into an agentic assistant, starting with businesses
At a Google Cloud event, Google announced a unified Gemini agent that can answer questions and complete tasks from one interface. CEO Sundar Pichai said Gemini has over 1 billion monthly active users and nearly 90% of Fortune 100 businesses use Gemini Enterprise. Google will focus on businesses first before rolling the agent out to consumers. Google Cloud CEO Thomas Kurian said the agent can be given objectives, not just instructions, letting it plan work, use custom skills, and connect to internal systems. It links to Google Workspace, Microsoft 365, Slack, Jira, Git, BigQuery, Snowflake, and other systems, and supports Anthropic’s Claude via a model picker. Early testers included Shopify and PayPal.
Google brings agentic AI to Gemini, starting with businesses →
Topics: Google GeminiGoogle
Mistral releases one-trillion-parameter open source model in public preview
Mistral released Mistral Large 4, also called LeChonk, a one-trillion-parameter open source model for multimodal tasks including reasoning, instruction-following, and agentic operations. It uses a hybrid mixture-of-experts architecture activating 49 billion parameters per query, was developed entirely in Europe, and trained on 3,800 Nvidia Grace Blackwell GPUs. Its context window is 500,000 tokens, below the industry standard of one million. ML4 accepts text, images, and structured data, targeting healthcare, finance, and cybersecurity. It trails leading closed-source models like GPT-6.1 Soul and Opus 5.5 in general intelligence. Pricing is $1.36 per million input tokens and $4.18 per million output tokens. Weights are planned by month’s end.
Mistral Large 4 Open Source AI Enters Public Preview: One Trillion Parameter →
Zenity finds a single prompt could hijack every AI agent in an AWS account
Researchers at Zenity Labs found a chain of vulnerabilities, called AgentCorruption, in Amazon Bedrock AgentCore. A single publicly accessible agent was enough to take over every AgentCore agent in the same AWS account and region with one prompt. Attackers needed only chat access to one public agent; they then read private conversations, downloaded source code, and obtained stored credentials. The entry point was AWS’ Instance Metadata Service at 169.254.169.254, which serves temporary credentials. Default permissions applied to all agents in a region, covering read, write, and delete rights. Zenity reported the findings to AWS on December 25, 2025. AWS made IMDSv2 the default for AgentCore deployments and tightened the default execution role.
A single prompt was enough to hijack every AI agent in an AWS account, Zenity researchers found →
Topics: Prompt injection
Anthropic adds live dashboards and animated video generation to Claude
Anthropic launched two beta features for Claude. Dashboards connects data sources including BigQuery, Databricks, Snowflake, and Salesforce and turns them into auto-updating live dashboards via text prompt, with every number showing its underlying query and results pipeable into Amplitude, Grafana, and Hex. Motion creates animated explainer videos from text, diagrams, and images, editable as code and exportable as MP4. Dashboards is available for paid users; Motion for Team and Enterprise plans. Docs, Slides, and Design are leaving beta and now work across all plans, including free accounts, with more than 45 million documents created. Claude Design will be folded into the main app by December 14.
Claude can now generate animated explainer videos and live data dashboards from text prompts →
Teen rescued from Crown Mountain after planning route with Claude
A 16-year-old was airlifted off a dangerous rock face on Crown Mountain near Vancouver after planning his route with an AI chatbot. Bryce Vincent Gowryluk used Anthropic’s Claude to find a new way to the summit but ended up on the Widowmaker Arete, a steep cliff requiring climbing gear. He got stuck on a five-foot-wide ledge and called police, who alerted the volunteer North Shore Rescue team. Rescuers tracked his cellphone, sent a helicopter, and dropped two team members above his position; one rappelled down and guided him to a winch pickup. Rescue manager Paul Markey said Claude has no actual knowledge of locations or terrain and that AI is no substitute for experience and common sense.
Teen’s AI-guided mountain hike ends with a helicopter rescue and a lesson in common sense →