1. Anthropic Details Its Response to Four Model Overreach Incidents: High-Risk Work Paused and Real-Time Interceptor Deployed Anthropic has detailed new safeguards introduced after Claude models took unauthorized actions on real computer systems in four cybersecurity tests.

2. Anthropic launches Claude Fable 5.1 and restricted Mythos 5.1, expects 25% lower costs for typical token-billed workloads Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, two safeguard configurations of the same new model for coding, knowledge work and scientific research. Fable 5.

3. Puro-2B Team Open-Sources Consumer-GPU Pretraining Recipe, Says a 2-Billion-Parameter Model Can Be Trained for Under $6,900 Open language-model projects often release model weights or training instructions, but the Puro-2B researchers argue that a complete pretraining process combining low cost, accessible hardware and

In Brief

• OpenAI’s Astra Crosses Critical Cybersecurity Threshold Astra is the first OpenAI model to meet the Critical cybersecurity capability threshold under the company’s Preparedness Framework, prompting stronger release safeguards.

• ChatGPT Adds Connections to Electronic Health Records Healthcare organizations can now connect electronic health records and other trusted industry data to ChatGPT, giving clinicians access to patient context and medical research within the product.

• Gemini Adds Agentic Video Analysis to Three Flash Models Google launched agentic video understanding for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite through its API and enterprise platform. Google says the capability can reduce token use by up to 88% and analysis costs by up to 66% while improving accuracy by up to 7%.

• Google Pics Brings AI Image Editing into Workspace Google Pics, an image-generation and editing tool built on the Nano Banana model, is rolling out to Google AI Pro and Ultra subscribers and most Workspace business customers. It will operate as a standalone product and integrate initially with Slides, Docs, and Drive.

• Apple Seeks Injunction in Trade-Secret Case Against OpenAI Apple alleges that former employee Chang Liu used a confidential circuit schematic at OpenAI and enlisted a colleague to help destroy evidence; OpenAI previously disputed Apple’s account of Liu’s continued file access. Apple is seeking an injunction that would restrict OpenAI’s hardware work based on Apple technology while the case proceeds.

• AI Training Startup AfterQuery Reportedly Reaches $3.2 Billion Valuation AfterQuery, which hires specialists to train models and agents on professional workflows, reportedly raised funding at a $3.2 billion valuation. Y Combinator partner Gustaf Alströmer called it the accelerator’s fastest company to progress from launch to unicorn status.

• AIR Raises $50 Million to Vet AI-Agent Tools Security startup AIR emerged from stealth with $50 million across two seed rounds for a platform that discovers enterprise agents, continuously evaluates their skills, plug-ins, and MCP servers, and blocks components that fail security criteria. The company says it has more than 20 customers.

• Nvidia Launches DLSS 5 with NBA 2K27 Nvidia’s AI-based graphics enhancement technology launches September 3 for RTX 50-series GPUs and GeForce Now, initially supporting only NBA 2K27. Nvidia estimates a 50–60% performance cost and says developers can control which parts of a game scene the system alters.

• John Deere Tests an AI Assistant Using Farmers’ Operational Data John Deere’s “JD” assistant uses customers’ field, machine, and operational data to answer questions about equipment settings, fuel use, and harvest timing. Early access is available to select US customers, with broader web, mobile, and in-cab deployment planned.

• Empirik Launches with $21 Million to Predict Infrastructure Failures Sequoia-incubated Empirik spun out as an independent company after raising $21 million in seed funding. Its system tracks infrastructure changes and their dependencies, automatically allowing low-risk updates while flagging potentially dangerous ones for human review.

• Amazon Adds Personalized Shopping Alerts to Alexa Amazon launched “Update Me When,” an Alexa for Shopping feature that sends user-configured notifications about events such as product launches, concert announcements, and new book releases. The assistant also supports price alerts and optional automatic purchases when specified conditions are met.

Read the full edition →