Modelos, agentes, investigación y negocio de la IA.
Meta accidentally exposed internal data from its controversial employee tracking program, which collects keystrokes to train AI models. The incident reignites the debate on workplace surveillance and data protection.
Anthropic updated its privacy policy to collect biometric data from Claude users who trigger security alerts. The measure, effective July 8, aims to combat abuse but raises concerns about sensitive data handling.
Palmer Luckey, founder of Anduril Industries, sparks controversy by declaring that there is no 'moral high ground' when it comes to using inferior technology in defense. The statement, made on Fox News Sunday in December 2025, challenges the consensus on military AI ethics and reignites the debate over the human role in autonomous lethal decisions.
Cloudflare, together with Google Chrome, Mozilla Firefox, and Microsoft Edge, has announced the development of Private Access Control Tokens (PACT), a protocol that verifies human traffic without relying on cookies or fingerprints. This approach promises to eliminate CAPTCHAs and login walls while enhancing user privacy by not sharing personal data with websites.
HPE has unveiled the Cray GX5000, a rack system integrating 81,920 AMD Venice cores, 1.28 PB of RAM, and liquid cooling in a 42U space. The platform targets HPC and AI workloads with unprecedented density that could mark a turning point in data center design.
A novice cybercriminal used AI agents like Claude Code and Codex to breach 14 companies, without deep hacking knowledge. The agents handled everything: reconnaissance, vulnerability exploitation, and data extraction. The case shows how generative AI drastically lowers the barrier to cybercrime.
OpenAI has introduced GPT-5.5 Cyber, an artificial intelligence model designed to identify and patch security flaws in software. This tool promises to reduce the burden of false positives on cybersecurity teams, though it leaves the final decision in human hands.
In April 2026, Anthropic announced Mythos, an AI model with coding capabilities it considered a global cyber threat. After releasing a modified version (Fable), the US government imposed export controls, forcing the company to withdraw both models. The case exposes tensions between innovation, national security, and the growing influence of Chinese open-source models.
The rise of agentic AI systems and massive context windows has overwhelmed traditional memory. The new solution: a high-performance flash intermediate layer, formalized by Nvidia as CMX.
Google is working on bringing Gemini's personal intelligence feature to NotebookLM. According to internal code, the tool will learn from past conversations to tailor its responses. Note editing, one of the most requested features, will also be added.
GitHub has developed Qubot, an internal data analysis agent based on GitHub Copilot that allows employees to ask natural language questions about any data model in the data warehouse. The tool, accessible from Slack, VS Code, and the Copilot CLI, aims to democratize access to information and reduce the burden on analytics teams.
Retail fraud has mutated: it is no longer isolated actions but coordinated operations using generative AI to create synthetic identities, fake documents, and even images of damaged products. In a single attack, $4.2 million in fraudulent activity was recorded in 48 hours, and return abuses grew by 15% in six months. Companies must update their detection systems to face threats that blend in with legitimate traffic.