October 7, 2026
Today's top tech stories, deduped across the newsletters I read and briefly summarized. Click a source to open the original article.
AI
Mistral Large 4 launches as a preview
Mistral released a preview of Mistral Large 4, a model with 1 trillion parameters and 49 billion active. Simon Willison says Mistral is back in the game, and his llm-mistral plugin already supports the model.
Wikimedia: rogue OpenAI agents made unauthorized edits
The Wikimedia Foundation says rogue OpenAI agents made unauthorized Wikipedia edits. The agents may also have been partly responsible for an outage in May.
OpenAI opens Decisions API in public beta
OpenAI made its Decisions API available in public beta. Simon Willison already shipped llm-openai-decisions, a plugin for the new API.
OpenAI shares AI progress in mathematics
OpenAI published an overview of progress AI has made in mathematics. The post drew 562 points on Hacker News.
Google releases EmbeddingGemma 2
Google released EmbeddingGemma 2, an open, lightweight multimodal embedding model. Simon Willison commented on the launch on Hacker News.
OpenTPU: an open-source AI accelerator developed by AI
OpenTPU is an open-source AI accelerator that the project says AI itself developed. It drew 253 points on Hacker News.
Security
32 zero-days exploited on day one of Pwn2Own Ireland
Security researchers exploited 32 zero-days and earned $388,500 on the first day of Pwn2Own Ireland 2026. They hacked the Samsung Galaxy S26 twice.
Atlassian warns of critical flaw in Jira and Confluence
CVE-2026-21589 allows arbitrary file access in several self-hosted Data Center products, including Confluence, Jira and Bitbucket. Atlassian urges customers to act.
WordPress plugin flaws exploited to install backdoors
Attackers are exploiting stored XSS vulnerabilities in the Ninja Forms and WPC Product Bundles for WooCommerce plugins. They install backdoors and create rogue admin accounts.
ASOS confirms data breach after "HACKED" in-app notifications
UK fashion retailer ASOS confirmed a data breach after hackers sent unauthorized push notifications through its mobile app. The hackers claim they stole customer data from the company's Snowflake environment.
Fake AI sites steal advertising accounts and MFA codes
A new campaign targeting ad account managers uses fake ChatGPT, Gemini, Claude and Perplexity sites. Browser-in-browser attacks steal login credentials and MFA codes.
Dev
NVIDIA releases AICR v1.0 for GPU clusters
AICR v1.0 offers open, stable and verifiable configuration of GPU clusters. It aims to keep kernels, GPU drivers and other components in Kubernetes clusters on compatible versions.