Modelos, agentes, investigación y negocio de la IA.
Trust in AI tools for security testing has collapsed: from 29% to 9% in one year. 78% of professionals report that automatic scanners miss critical vulnerabilities. Hybrid models, combining automation with human oversight, emerge as the preferred alternative.
Oracle has eliminated 21,000 jobs in the last fiscal year, according to a regulatory filing. The company directly links these cuts to AI adoption and plans to finance its AI investments through debt issuance.
Superhuman, the company behind the writing assistant Grammarly, has acquired GPTZero, a startup specializing in detecting AI-generated text. The deal, announced in June 2026, aims to integrate AI detection capabilities into its productivity tool ecosystem, amid growing concerns about the authenticity of digital content.
A technical article from Towards Data Science describes how to run three different LLMs on a single 8 GB GPU using layer multiplexing in C++ and admission control. This technique overcomes VRAM limitations and enables multi-agent systems on modest hardware.
Baseten, an AI model deployment infrastructure company, has announced a $1.5 billion Series F round that values it at up to $13 billion. The investment, the largest from Australian Blackbird VC, reflects growing demand for scalable inference platforms.
The intelligence chiefs of the Five Eyes countries issued a joint warning: artificial intelligence is shortening the time between vulnerability detection and exploitation, potentially turning security incidents into operational and financial crises. They urge business leaders to urgently adopt basic cybersecurity measures.
A team from the Vesuvius Challenge has used artificial intelligence to read a carbonized Herculaneum papyrus that cannot be physically opened. The text reveals Stoic philosophy and may belong to the philosopher Chrysippus.
California has launched an AI-powered unemployment tracker to measure the impact of automation on employment. The tool, developed with the University of California, provides monthly data and aims to replace speculation with evidence.
Liquid AI has unveiled LFM2.5-230M, an ultralight foundational language model that promises to transform data extraction on the edge. With performance surpassing models four times larger and the ability to run locally on devices like the Raspberry Pi, it signals a paradigm shift toward architectural efficiency over brute scaling.
The race for speed in AI code generation gives way to the challenge of trust. Experts warn that the next bottleneck is not the ability to generate code, but to verify, govern, and deploy it reliably.
A team from the Vesuvius Challenge has managed to read from start to finish the preserved text of a Herculaneum papyrus carbonized by the eruption of Vesuvius. The feat, combining high-resolution X-ray microtomography and AI models, promises to unlock hundreds of scrolls from the only intact library of antiquity.
The White House has asked OpenAI to delay the public launch of GPT-5.6 due to security concerns. In response, the company plans to share the model only with a small group of partners. This move marks a milestone in artificial intelligence regulation.