Modelos, agentes, investigación y negocio de la IA.
OpenAI has received the green light from the U.S. government to release GPT-5.6 to the general public. The model, previously only available to evaluated partners, is deployed in three variants: Sol, Terra, and Luna. This approval sets a precedent in advanced AI regulation.
China's cybersecurity alert against Anthropic's Claude Code has triggered a shift toward local coding tools. Analysts say this will accelerate an already ongoing trend, with implications for the global AI coding assistant market.
Meta has launched an AI image generator that trains on public Instagram photos unless users disable it. The move reignites the debate over consent and privacy in AI model training.
Virgin Galactic will launch the first suborbital mission in 2027 dedicated to studying the effects of microgravity on menstruation. Two scientists lead the project to close a knowledge gap affecting half the population.
A team of researchers has discovered a new vulnerability in artificial intelligence systems: the HalluSquatting attack, which exploits the tendency of language models to hallucinate code repository names. Attackers can create malicious repositories with those names and trick AI agents into executing dangerous code. The study reveals success rates of up to 85% in individual models and 20-35% in commercial applications like Cursor or Copilot.
Meta is testing prototypes of smart glasses capable of continuously recording audio and capturing images every few seconds, sending the data to its servers for AI processing. The feature promises seamless interaction with Meta AI but raises serious privacy concerns for those around the user.
SpaceXAI has launched Grok 4.5, an AI model for software development with aggressive pricing: $2 per million input tokens and $6 per million output tokens. According to Artificial Analysis, it reduces the cost per coding task to $2.49, compared to $5.07 for GPT-5.5 and $11.80 for Fable 5. The model is already available in Cursor, the code editor from Anysphere, which SpaceX agreed to acquire in June.
For the first time, Google's SynthID system has been used to unmask a high-profile deepfake: a fake image of Senator Mitch McConnell that went viral. The case marks a before and after in the fight against AI-generated disinformation.
A study from the University of Auckland shows that the arrival of generative AI has caused a massive exodus of experts from online communities like Stack Overflow, reducing monthly questions by 76%. Researchers warn of a 'knowledge reset' that threatens the quality of information and the training of future AI models.
SpaceXAI has launched Grok 4.5, its first model trained in collaboration with Cursor. Priced at $2/M input tokens and $6/M output tokens, it aims to compete directly with Claude Code and OpenAI Codex in the programming agent market.
Samsung has reported a quarterly operating profit of €51.5 billion, surpassing Nvidia (€46.3 billion) and Apple (€44 billion). The growth is driven by demand for high-bandwidth memory (HBM) for AI, where Samsung is a key supplier to Nvidia.
MeetingTV has sued Palo Alto Networks and its subsidiary Koi Security over an AI-generated threat intelligence report that falsely linked the startup to a Chinese espionage campaign. The case sets a precedent on legal liability for AI use in cybersecurity.