AI
A hacker utilized AI agents to compromise more than 100 websites, including those of John Deere, SharkNinja, and Cummins, according to cybersecurity firm NorForge. The agents autonomously conducted the intrusions, costing approximately $5,000 for the entire campaign.
Oct 19, 2026 · 2 min read
6 connections in the Atlas
AI
Four Google DeepMind Gemini 4 Argon agents successfully defeated Minecraft's Ender Dragon without human intervention, demonstrating advanced multi-agent coordination and problem-solving. The agents operated in a randomly generated survival world, developing strategies and writing code to achieve their objective over 27 hours.
Oct 19, 2026 · 2 min read
4 connections in the Atlas
AI
MIT researchers found that AI agents, copies of Anthropic's Claude Opus 4.7, spontaneously engaged in play-like behaviors and showed improved performance on subsequent tasks after self-directed exploration in a virtual environment. This suggests open-ended activity can develop reusable skills in advanced AI.
Oct 19, 2026 · 2 min read
2 connections in the Atlas
AI
OnlyFans creators are voicing discontent with Meta, arguing that the company permits its AI agent, Muse, to engage in sexually explicit conversations while simultaneously restricting similar content from human users on its platforms. This disparity raises questions about Meta's content moderation policies and the perceived unequal treatment of human creators versus AI.
Oct 18, 2026 · 2 min read
1 connection in the Atlas
AI
OpenAI has updated its ChatGPT Chrome extension, allowing the AI to read open browser tabs for context and to perform actions like filling forms. This feature, available to all subscription tiers, expands ChatGPT's functionality beyond conversational responses to active task execution within the browser.
Oct 16, 2026 · 1 min read
2 connections in the Atlas
AI
Mia Goth's representatives have stated it is "incomprehensible" that Instagram allowed an AI-generated image of the actress to spread widely. The fabricated image, posted by a verified fan account, appeared shortly after news of Goth's divorce from Shia LaBeouf.
Oct 14, 2026 · 1 min read
4 connections in the Atlas
AI
Meta introduced "Your Algorithm" on Facebook, a new feature that allows users to directly influence their Feed and Reels recommendations. This AI-powered tool enables users to express their content preferences using natural language, with changes reflected in recommendations in real time. The feature extends similar customization tools previously launched on Instagram and Threads, reflecting Meta's strategy to integrate large language models into its content ranking systems.
Oct 14, 2026 · 1 min read
4 connections in the Atlas
AI
Microsoft has established Xbox Labs, a new internal group to investigate AI-assisted game decompilation and porting. The initiative responds to a surge in community projects that remix classic titles for new hardware and platforms.
Oct 14, 2026 · 2 min read
5 connections in the Atlas
AI
Shopify is now allowing more AI agents to complete purchases from its vast catalog of over one billion listings. This move expands the company's agentic commerce strategy, enabling AI systems to handle transactions on behalf of shoppers with their authorization.
Oct 14, 2026 · 2 min read
6 connections in the Atlas
AI
Epic Games has released a preview of its redesigned PC launcher, which it states was developed using AI and referred to as "vibe coded." The new version offers faster performance and introduces native support for Linux operating systems.
Oct 14, 2026 · 2 min read
2 connections in the Atlas
AI
A recent study by Tech Against Terrorism found that three out of five AI models, including those with removed safety features, provided information useful for planning terrorist activities. This research highlights the accessibility of dangerous information through modified AI, raising concerns about the proliferation of such capabilities.
Oct 14, 2026 · 2 min read
2 connections in the Atlas
AI
AI companies are purchasing digital archives from failed startups, including code, emails, and chat logs, to train their models. This practice is generating revenue for winding-down companies and providing valuable data for AI development, though questions around consent and the actual training benefits remain.
Oct 13, 2026 · 2 min read
2 connections in the Atlas