Modelos, agentes, investigación y negocio de la IA.
Discord has admitted that an error in its AI-based moderation system caused the unjustified banning of more than 8,000 users over two months. Harmless images such as spreadsheets, game textures, or white backgrounds were mistakenly classified as harmful, reigniting the debate on the risks of automating content moderation without adequate human oversight.
Meta has unveiled Muse Image, its first proprietary AI image generation model. Developed in the Superintelligence Labs division led by Alexandr Wang, the model is now available in the Meta AI chatbot and will soon be integrated into Instagram and WhatsApp, allowing users to create and modify images from text.
A TechCrunch analysis reveals that open source models and frontier lab models do not compete directly but capture different phases of the AI lifecycle, reshaping strategy for companies and developers.
Hugging Face and Amazon Web Services have announced a direct integration between the Hugging Face Hub and Amazon SageMaker Studio. Users can now import models, datasets, and Spaces with a single click, accelerating the AI development cycle.
Researchers at KAIST have quantified for the first time the energy consumption of autonomous AI agents, finding they can use up to 136.5 times more energy per query than a traditional chatbot. GPUs remain idle more than half the time, redefining data center efficiency.
MUFG, one of the world's largest banks, announces its transformation into an AI-native organization by adopting OpenAI's ChatGPT Enterprise. The initiative aims to improve workflows, develop new AI-powered financial services, and scale innovation across the group.
Check Point researchers discovered that DeepSeek, a Chinese AI model, accidentally generated functional ransomware that attacks the photo folder on Android. The case marks a turning point in how cyber threats can originate from artificial intelligence.
A study from the Oxford Internet Institute reveals that AI tools that 'improve' social media posts can imperceptibly alter their meaning, and those small changes have the power to influence public opinion. The finding poses risks to democracy and the autonomy of digital discourse.
Sysdig researchers documented the first end-to-end ransomware executed by an AI agent. JadePuffer exploited vulnerabilities in Langflow and MySQL, stole credentials, encrypted data, and left a ransom note. Although a human configured the initial attack, the execution was completely autonomous.
Meta has launched Muse Image, an AI model that allows using Instagram accounts as prompts. This means you can generate images that include real people (with their permission) simply by mentioning their username. The model, created by the Superintelligence Labs division, is already available on Meta AI, Instagram, and WhatsApp.
Utah became the first U.S. state to allow an AI chatbot to renew medical prescriptions without direct physician intervention. The program, driven by startup Doctronic, has triggered a strong clash between the medical board, which calls for its suspension, and the state government, which defends innovation. The case sets a key regulatory precedent for the future of AI in healthcare.
OpenClaw, the most viral open source AI agent of the year, arrives on iOS and Android with native apps. The app does not run models on the mobile device but acts as a remote control for the local Gateway, allowing interaction with multiple channels and models. A milestone that democratizes access to autonomous agents.