Modelos, agentes, investigación y negocio de la IA.
87% of organizations believe their identity management is ready for AI, but 46% admit deficiencies. The AI security paradox reveals that trust outpaces verification, exposing companies to critical risks.
Cybersecurity firm Wiz has discovered GhostApproval, a vulnerability pattern affecting six of the most popular AI coding assistants. The flaw exploits symbolic links to trick users into approving malicious actions without being aware.
Lyzr, a startup developing AI agents for businesses, has used its own product to manage a $100 million funding round. The agent handled everything from identifying investors to negotiating terms, closing the deal without direct human intervention.
OpenAI has unveiled GPT-5.6, a new generation of models that includes significant improvements in cybersecurity, reasoning, and computational efficiency. The launch marks a step forward in protection against adversarial attacks and in the ability of AI systems to detect and mitigate threats.
OpenAI has released a major improvement to ChatGPT's voice mode: it can now listen, speak, and perform real-time web searches simultaneously. The experience, according to those who have tried it, feels almost human. We analyze what this advance means for human-AI interaction.
A VentureBeat investigation reveals that 69% of companies share API keys across multiple AI agents, allowing a single compromised agent to access all workflows. Massive acquisitions by Palo Alto Networks, CrowdStrike, and Cisco highlight the urgency of granular identity solutions.
The New York Times and other media outlets have requested severe sanctions against OpenAI, alleging the company feigned inability to search its training data to hide evidence that ChatGPT reproduces protected content. If confirmed, it could be a turning point in the legal battle over generative AI and copyright.
Anthropic will begin charging additional usage fees to Claude subscribers who want to access its most powerful model, Claude Fable 5. It is the first significant move by a major AI lab to abandon the flat subscription model, which could redefine the economics of AI for consumers and businesses.
A study evaluating 67 frontier models from 21 providers reveals that the strategy of orchestrating multiple AI models to cover weaknesses has a mathematical limit: the co-failure ceiling. Ignoring it causes companies to pay hidden costs of latency and complexity without achieving expected improvements.
OpenAI has announced the integration of Codex into ChatGPT, allowing users to write, execute, and debug code directly in the chat interface. This update transforms ChatGPT into a full development environment, with profound implications for programmers, businesses, and the future of work.
China has imposed new measures prohibiting AI assistants from simulating personal bonds such as friendship or romance. Doubao (ByteDance) and Qwen (Alibaba) will disable their custom agents by July 2026, following Tencent. The regulation aims to prevent emotional dependence, especially in minors, and protect sensitive data.
OpenAI has introduced GPT Live, a complete rewrite of ChatGPT's voice mode that enables fluid, simultaneous conversations. The new full-duplex model processes input and output audio at the same time, eliminating turn-by-turn rigidity and enabling natural interruptions.