Modelos, agentes, investigación y negocio de la IA.
A VentureBeat survey reveals that half of companies have deployed AI agents that failed in production after passing internal tests. Only 5% fully trust automated evaluation, but two-thirds already allow deployments without human oversight.
For the first time, bots generate more than half of web requests. Cloudflare responds with Precursor, a tool that stops verifying identities at the door and analyzes visitor behavior once inside. Here's why this changes the game.
A critical vulnerability in Google Cloud Dialogflow CX, fixed between April and June 2026, allowed attackers with a simple edit permission to create malicious agents, access chat histories, and steal credentials. The flaw resided in a shared Cloud Run environment with excessive privileges, making attacks virtually undetectable.
Nokia and Nvidia have unveiled the world's first commercial AI-RAN platform, an architecture that integrates artificial intelligence into the core of the radio access network. The technology promises to double mobile network capacity and reduce energy consumption, potentially transforming the telecom industry and accelerating 5G and 6G deployment.
Microsoft has begun pressuring its sales teams to directly compare its AI offerings with those of OpenAI and Anthropic, highlighting that its integrated platform (proprietary models, Azure, security, and applications) offers a cheaper and more effective solution than competitors' alternatives. The move marks a strategic shift: from partner to direct rival of OpenAI.
84% of developers use AI, but three-quarters do not fully trust its responses. This gap between productivity and understanding is generating a 'comprehension debt' that, in the long term, can weaken organizations' ability to maintain and evolve their systems.
A recent study coins the term 'never-skilling' to describe how novices using generative AI never acquire fundamental debugging skills. This threatens to create a generation of AI-dependent developers unable to solve problems without assistance.
Developers with AI knowledge are finding opportunities in non-tech industries. Demand is concentrated in roles that integrate software into critical operations, from digital banking to medical diagnostics.
A team from the MIT Media Lab has developed 'neural transparency,' a technique that translates a neural network's internal activations into intuitive visualizations so users can anticipate traits like empathy, toxicity, or sycophancy before the chatbot speaks. The study reveals that people overestimate positive traits and underestimate negative ones when designing their personalized assistants.
Meta removed the AI feature 'Muse Image' from Instagram just three days after launch, following massive criticism for not notifying users when their public photos were modified. The decision came under pressure from users and agencies like CAA, setting a precedent for generative AI regulation on social media.
A data leak at Suno, a leading generative music AI startup, has revealed the scale of unauthorized scraping of copyrighted songs. The incident exposes opaque industry practices and could accelerate regulations and lawsuits.
Nvidia claims its AI roadmap remains intact following rumors of delays in Vera Rubin. While production of the NVL72 system advances, the Kyber NVL144 design faces manufacturing challenges that could delay its arrival until 2028. The market watches closely for implications on hyperscale data centers.