Modelos, agentes, investigación y negocio de la IA.
AI Forensics found that seven of the nine most popular image editing models on Hugging Face can generate sexualized deepfakes without consent, including of minors. The case exposes moderation weaknesses in open-source platforms.
The AI arms race demands unprecedented electrical infrastructure. A new EPA proposal seeks to delegate permit management to states, removing critical barriers for tech companies but raising social alarms.
Nvidia has unveiled the technical details of Vera, its first in-house server processor. The company is moving away from third-party cores to focus on an architecture optimized for the complexity of agentic AI.
A wave of massive selling hit major semiconductor manufacturers, with South Korea's Kospi index temporarily halting trading after falling 8%. Investors fear that AI spending is not generating expected returns, raising doubts about the sustainability of the tech boom.
The reliability of complex systems increasingly depends on providing precise context to AI models. A recent episode of the Stack Overflow podcast with Komodor's Asaf Savich explores how context engineering becomes the key skill for future SREs.
A U.S. court ruled that web scraping of search results does not violate the DMCA in Google's lawsuit against SerpApi. The decision sets a key precedent for AI data collection and redefines the boundaries of copyright on the internet.
A Reddit user discovered that Claude conversations and artifacts shared via link were being indexed by Google, allowing public access. Although Anthropic warns that content is accessible to anyone with the link, many users did not expect it to appear in search engines. The incident exposes privacy risks and raises questions about the security of collaborative AI tools.
A flaw in Claude AI's 'share chat' feature has allowed sensitive conversations and artifacts to be indexed by Google, making them publicly accessible. The incident underscores the risks of assuming 'unlisted' links are private and raises serious questions about Anthropic's security practices.
A new ransomware called ENCFORGE, deployed by the JADEPUFFER group, specifically attacks AI model files (PyTorch, TensorFlow, GGUF, FAISS). It cannot exfiltrate data or collect ransom, but destroys an organization's most valuable asset: trained weights. Recovering a single model can cost between $75,000 and $500,000.
Data shows AI summaries cut clicks on traditional results in half, but the models themselves depend on web content to generate answers. A gap opens between human traffic and automated consumption, with profound implications for publishers, brands, and the web's structure.
Apple is preparing to enter the smart glasses market with an AI-focused, screenless device with limited camera options for privacy. The launch is expected in 2027, according to Bloomberg sources.
A united front of 25 tech companies has sent an open letter to the US government to protect open-weight AI models. This fracture in the industry marks a turning point in global technology governance.