Une erreur de migration a ouvert les archives de DigiCert à des comptes créés sur Internet : données clients et pièces d'identité consultées. Ce que l'on sait.

Le post DigiCert : une erreur de migration expose les données et pièces d’identité de clients a été publié sur IT-Connect.

Après les supermarchés et l'EFS, la CISA et le FBI alertent sur les attaques contre les systèmes de froid industriels. Voici leurs recommandations.

Le post Après l’Europe, le FBI et la CISA alertent sur les cyberattaques contre les systèmes de réfrigération a été publié sur IT-Connect.

La faille *sshole (CVE-2026-123158) ouvre en SSH, avec un mot de passe trop long, des comptes Linux sans mot de passe. Alpine Linux est le premier concerné.

Le post *sshole : sur Alpine Linux, un mot de passe trop long peut mener à un accès root via SSH a été publié sur IT-Connect.

OpenAI lance GPT-6 Cyber, réservé à Daybreak Red : 948 exploits Linux, 81 failles graves dans Firefox et un modèle entraîné à rester dans son périmètre.

Le post OpenAI lance GPT-6 Cyber : son modèle IA taillé pour trouver et exploiter des failles a été publié sur IT-Connect.

Supprimez les pubs de l'accueil de votre box Android TV avec CleanYourTV et Claude Code, ou une app gratuite : ADB, prérequis, thèmes et précautions.

Le post CleanYourTV : comment nettoyer une box Android TV avec l’IA (sans root) a été publié sur IT-Connect.

GPT-6 Cyber is OpenAI’s model for finding software vulnerabilities, testing their impact, and helping develop fixes. OpenAI says it uncovered 81 previously unknown high-severity Firefox flaws and developed working exploits for 948 Linux kernel vulnerabilities. The model is initially available through the API and Codex to approved Daybreak Red customers.

Source

Mistral Studio lets businesses build AI agents, workflows, and apps connected to their data and systems. Its new managed agents can run code and delegate tasks, while apps add hosting and access controls for employee-facing tools. The platform supports cloud, private cloud, and on-premises deployments.

Source

Gemini Omni 1.2 Flash generates videos from prompts and reference material across Google’s developer and creative tools. The update adds native clips up to 15 seconds, conversational editing, and stronger consistency, while cutting the 720p price to $0.07 per second.

Source

Inkling 2 Small is Thinking Machines Lab’s mixture-of-experts model, with 594 billion total parameters and 17 billion active. It is available now for inference and fine-tuning on Tinker, and supports image inputs and adjustable reasoning effort. The company plans to release its weights under Apache 2.0 next week, after external safety testing.

Source

Perplexity Computer is an AI service that uses agents to complete tasks; its new Teammates feature turns those agents into persistent workers for ongoing projects. They can take on different roles, reuse existing projects and memory, and coordinate through Slack. Teammates are available on the web to Perplexity Max subscribers, a plan priced at $200 per month.

Source

A PowerShell MCP server lets Claude Desktop call specific Windows functions instead of generating commands for an administrator to run. A guide builds one with PSMCP and four read-only checks for disk space, services, system events, and pending reboots. But the early-stage module has defects that can break tool discovery or leave calls hanging, and the server runs with the signed-in user’s permissions.

Source

Matthew Berman reviews a wave of AI model releases, from Moonshot’s Kimi K3.1 Pro to new computer-use, image, video, and science models. The roundup highlights open weights, faster generation, and smaller models designed for specific tasks. It also covers tools for document retrieval, voice-driven actions, and coding.

Source

Microsoft is overhauling Microsoft 365 Education, adding a free Windows-only student plan and a premium A7 bundle with Copilot and advanced security. Schools can choose student and faculty licenses with or without Teams. Pricing changes begin at renewals from May 1, 2027; most plan retirements start at renewals from January 1, 2028.

Source

Cloudflare has launched the Trusted Agent Delegation coalition to develop shared standards for AI agents acting on the web. It is also rolling out site controls that let customers see agent traffic, set access rules, and identify which agent made a login or sign-up request. These changes aim to help websites manage automated activity without blocking every user of a shared agent.

Source

Google will start replacing Gemini Gems—custom assistants built around saved instructions—with Skills for personal accounts held by users 18 and older on November 17, 2026. Gems will become read-only that day and disappear on December 17, so users with attached files should save them before the cutoff: some will not migrate. Skills can be called into a chat and can run automatically when their instructions match a prompt.

Source

openSUSE Leap is a Linux distribution for servers and desktops. With Leap 16.1, its Agama installer lets users choose between a conventional writable system and an immutable system with a read-only root and transactional updates. The change folds Leap Micro into Leap; administrators choosing a mode should note that switching later requires a fresh install.

Source

Google’s Gemini Canvas workspace helps users create interactive tools with Gemini. Now they can upload a photo of a sketch and ask Gemini to code it, then refine the result in a live preview.

Source

ERNIE 5.5 Flash Base is a language model checkpoint for teams that want to adapt a model for their own tasks, rather than use a ready-made chatbot. Baidu has released its weights with 280 billion parameters, of which 14 billion are active per token. The checkpoint is available in FP8 and BF16 formats, but has not been fine-tuned or instruction-tuned.

Source

ChatGPT’s optional Lockdown Mode is designed to reduce the risk of sensitive data escaping through prompt injection. Researchers disclosed a now-patched flaw in HTML previews that could let malicious content bypass the protection and silently exfiltrate chats from the desktop app. OpenAI fixed the issue on September 18, 2026.

Source