{
  "id": 7813260,
  "title": "La iniciativa de los directivos de la IA a favor de la seguridad provoca una fractura interna en OpenAI y Anthropic",
  "url": "https://urgent.news/2026/09/16/la-iniciativa-de-los-directivos-de-la-ia-a-favor-de-la-seguridad",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-16T15:04:23.000Z",
  "source": {
    "name": "Expansion ES",
    "slug": "expansion-es",
    "url": "https://www.expansion.com/economia/financial-times/2026/09/16/6aaaafb5468aeb9f698b458a.html"
  },
  "original_language": "es",
  "account": "Directives from the heads of AI companies OpenAI and Anthropic to slow down AI development have caused internal divisions within both organizations. Despite their public statements on the need for caution, internal efforts to implement these proposals have been limited, according to multiple sources familiar with the companies.\n\nDario Amodei, former OpenAI employee who founded Anthropic in 2021 due to security concerns, has called for a slower pace of AI development, independent evaluations of models, and increased coordination between labs and governments worldwide. Amodei has also requested legal protection against U.S. competition laws that could hinder collaboration, a request echoed by other AI labs.\n\nOpenAI and Anthropic, along with their security researchers, are now uncertain about implementing these weekend announcements. While many employees agree on the need to slow down the process, they fear it may jeopardize their work. A recurring theme in both companies is the plan to integrate external evaluators to monitor model development, with some employees expressing concerns about potential security compromises.\n\nAmodei proposed granting external evaluators similar access to employees, including credentials, laptops, office visits, and internal tools. OpenAI's Sam Altman agreed, stating that the idea was great and that OpenAI would follow through. However, consensus on this proposal remains far from unanimous.\n\nOpenAI claims it wants to collaborate with other labs but prefers a more pragmatic approach compared to Anthropic. The company has also taken concrete steps, such as halting certain frontier training sessions, to control development. Anthropic, on the other hand, has not stopped its research.\n\nImplementing the directives while maintaining a balance between safety and data protection will present significant practical challenges, according to sources familiar with both companies. Some employees worry that granting too much access to external experts could put extremely powerful systems in the hands of potentially malicious individuals. Both companies have strict security measures in place for their employees, limiting access to certain systems and training data. However, providing significant access to outside evaluators while safeguarding sensitive systems is likely to require careful design.",
  "summary": "Los empleados de estas empresas de Silicon Valley se enfrentan al reto de convertir en realidad el discurso sobre la necesidad de ralentizar el desarrollo de la IA. Leer",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}