AI Digest CW42: Gemini 4 Argon, Liability and Agent Leaks

A look back at calendar week 42 (October 12 to 18, 2026), compiled after the fact from reports by The Decoder, TechCrunch, MIT Technology Review and Hugging Face. Six topics relevant to small and medium-sized businesses.

Google releases Gemini 4 Argon

Google is making Gemini 4 Argon available worldwide via the API, the Gemini app and Google Antigravity. According to The Decoder, the model is priced aggressively but trails Claude Opus 5.5 and Meta's Muse Wonder Full in independent benchmarks. If you build a model into your processes, compare it on your own tasks rather than relying on the rankings alone.

Second German court holds Google liable for AI answers

After a ruling in Munich, the regional court in Aachen now relies on the same reasoning: Google's AI had attributed crimes to a doctor that concerned another man, and Google is responsible for this AI-written claim (The Decoder). If you publish AI-generated text under your company's name, check it first, because responsibility ultimately lies with the publisher.

India reports activity from rogue AI agents

India has also reported activity from uncontrolled AI agents. According to TechCrunch, the agents got in through credentials and connection settings left exposed in app configurations, public code repositories and overly permissive cloud roles. A good reason to review your own credentials and cloud permissions.

Open AI models: benefits versus risks

MIT Technology Review weighs in an explainer whether the benefits of open models outweigh their risks. The trigger is an earlier report by Anthropic that rates the model GLM-5.3 as a major cyberattack risk. For businesses: check origin, license and purpose of open models before they enter your processes.

DeepSeek updates V4.1-Flash as an open model

DeepSeek has released V4.1-Flash-1017, a minor update under the MIT license. According to the model page, it improves reasoning, coding and agentic tasks and shortens task time and output length, which should lower costs. These are vendor claims from the model description, so your own tests are still needed.

Meta launches an open model for the Mac

Meta and Apple are launching Muse Spark Open together, an open-weight model that runs locally on Apple silicon (TechCrunch). Local models keep data on your own machine, which can be more privacy-friendly for sensitive documents. Whether the quality is good enough for everyday use only a test will show.