Modelos, agentes, investigación y negocio de la IA.
Apple has released iOS, iPadOS, and macOS 26.5.2 updates ahead of schedule, incorporating security patches originally planned for version 26.6. The decision responds to an increase in AI-based threats exploiting zero-day vulnerabilities.
Meta hired hundreds of workers to pose as teenagers and test rival chatbots like Gemini and ChatGPT on high-risk topics. The initiative aims to improve the safety of its own models but raises ethical questions about privacy and manipulation.
DeepSeek has released DSpark, a speculative decoding framework that accelerates large language model inference by up to 85%. Under the MIT license, it allows developers and companies to reduce deployment costs without sacrificing quality.
Google Cloud has announced the integration of SandboxAQ's quantum models into its marketplace, offering specialized models for numerical data and scientific equations. This move recognizes the limitations of traditional LLMs in scientific tasks and opens new possibilities in chemistry, pharmaceuticals, and materials.
OpenAI has announced the preview launch of the GPT-5.6 series, comprising three models: Sol (flagship), Terra (balanced), and Luna (fast and economical). Terra matches GPT-5.5 at half the cost, and Luna offers strong capability at the lowest price. The preview is limited at the request of the U.S. government.
The US government has asked OpenAI to restrict the launch of GPT 5.6 to a small group of approved partners due to its advanced capabilities. This measure, following the export control order against Anthropic, marks a new chapter in frontier AI regulation.
The Trump administration has asked OpenAI to release its next model, GPT 5.6, in a restricted manner, approving clients one by one. This marks the first time the government has directly intervened in the deployment of an OpenAI model, following precedents like the withdrawal of Anthropic's Claude Fable 5.
After a decade of development, confidential computing finds its killer use case in agentic artificial intelligence. At the Confidential Computing Summit, experts detail how Trusted Execution Environments (TEE) and hardware attestation provide the security that autonomous agents need to operate without exposing sensitive data.
Vercel has added adapters for Deep Agents and OpenCode to its AI SDK Harness, allowing developers to run these code agents in a unified sandbox environment. This makes it easy to switch between runtimes without modifying application code, accelerating AI-driven development.
In a Los Angeles arson trial, prosecutors presented ChatGPT conversations as evidence. The case could set a precedent for using AI assistant data in criminal proceedings.
A landmark ruling in Germany establishes that Google is legally responsible for errors in its AI-generated summaries. The decision, backed by security experts like Bruce Schneier, could transform how companies deploy AI assistants and redefines the concept of 'agent' in digital law.
The Bank for International Settlements (BIS) has warned that an AI investment bubble could burst and trigger a credit crisis of similar proportions to 2008. In its annual report, the institution points out that 'circular financing' and inflated expectations create systemic risks that regulators must address.