He wanted to make sure the decree would not become a barrier to progress.
En los primeros días de junio, el presidente Trump firmó una orden ejecutiva que invita a las grandes empresas de inteligencia artificial a compartir voluntariamente sus modelos más avanzados con el gobierno treinta días antes de su lanzamiento público. La medida nació de una inquietud genuina: que sistemas como Mythos, de Anthropic, pudieran amplificar ciberataques a una velocidad sin precedentes. En el fondo, el decreto refleja una tensión tan antigua como la tecnología misma: la búsqueda del equilibrio entre el impulso de innovar y la necesidad de proteger, entre la autonomía de la industria y la responsabilidad del Estado.
- El modelo Mythos de Anthropic encendió las alarmas en Washington al demostrar una capacidad sin precedentes para explotar vulnerabilidades de ciberseguridad, convirtiendo la supervisión gubernamental en una urgencia política.
- La Casa Blanca llegó a posponer abruptamente la firma de la orden —con ejecutivos ya en camino— porque Trump temía que el texto original obstaculizara el desarrollo de la IA, revelando la fragilidad del consenso interno.
- La industria negoció con fuerza: la ventana de revisión se redujo de noventa a treinta días, y el decreto quedó blindado con lenguaje explícito que prohíbe licencias obligatorias o autorizaciones previas del gobierno.
- Anthropic, la misma empresa cuyo modelo detonó la iniciativa y que el Pentágono había señalado como riesgo en la cadena de suministro, terminó participando en la redacción del decreto, ilustrando la paradoja de regular lo que uno mismo crea.
- El marco voluntario que emergió busca formalizar lo que ya ocurría en la práctica, preservando la autonomía de la industria mientras establece un mecanismo de revisión de seguridad antes de que las capacidades más peligrosas lleguen al público.
A principios de junio, el presidente Trump firmó una orden ejecutiva que otorga al gobierno federal acceso anticipado a los modelos de inteligencia artificial más poderosos antes de su lanzamiento público. El objetivo declarado era evaluar riesgos de ciberseguridad y proteger infraestructuras críticas de amenazas que los sistemas de IA avanzados podrían habilitar.
El detonante inmediato fue Mythos, el nuevo modelo de Anthropic, cuya capacidad para explotar vulnerabilidades de seguridad a una velocidad sin precedentes generó alarma tanto en Washington como en Wall Street. La administración argumentó que un período de revisión previo daría tiempo al gobierno para reforzar sus defensas antes de que esas capacidades quedaran al alcance de todos.
La versión final del decreto solicita a las empresas que compartan voluntariamente sus modelos treinta días antes del lanzamiento, una reducción significativa respecto a los noventa días propuestos inicialmente. La industria había advertido que plazos más largos podían frenar la innovación, y una reunión entre asesores de Trump, miembros del gabinete y líderes tecnológicos terminó de inclinar la balanza. La orden también incluye lenguaje explícito que prohíbe interpretar el requisito de divulgación como autorización para imponer licencias obligatorias o permisos previos, una salvaguarda que la industria exigió desde el principio.
El camino hasta la firma no fue sencillo. Trump había planeado firmar el decreto el 20 de mayo, pero lo pospuso abruptamente —con algunos ejecutivos ya en camino a la Casa Blanca— porque consideraba que ciertos aspectos podían obstaculizar el desarrollo de la IA. El texto volvió a revisarse antes de alcanzar su forma definitiva.
Lo que hace singular este episodio es la participación directa de la industria en la redacción de la norma que la regula. OpenAI y Microsoft celebraron la medida como un paso que concilia seguridad e innovación. Anthropic, cuyo modelo fue el catalizador de todo el proceso y que el Pentágono había señalado previamente como riesgo en la cadena de suministro, también contribuyó a dar forma al decreto. El resultado es un marco voluntario que busca sistematizar lo que ya ocurría de manera informal, sin convertir la supervisión gubernamental en un control obligatorio.
On a Tuesday in early June, President Donald Trump signed an executive order designed to give the federal government early access to the most powerful artificial intelligence models before they reach the public. The stated purpose was straightforward: to evaluate cybersecurity risks and protect critical infrastructure from threats that advanced AI systems might enable.
The timing of the order reflected genuine anxiety in Washington and on Wall Street. Anthropic's new model, called Mythos, had triggered particular concern because researchers claimed it could exploit cybersecurity vulnerabilities at an unprecedented pace. Experts had warned that sophisticated AI systems could amplify cyberattacks in ways that were difficult to predict or defend against. An early review period, the administration reasoned, would give the government time to shore up its defenses before dangerous capabilities became widely available.
The final version of the order asked AI companies to voluntarily share new models with advanced cybersecurity capabilities thirty days before making them available to other partners. This represented a significant retreat from an earlier draft that had proposed a ninety-day review window. The shift mattered because the AI industry had made clear to the White House that models were evolving so rapidly that longer delays could slow innovation meaningfully. A Monday meeting between Trump's senior advisors, cabinet officials, and industry leaders had helped push through the shorter timeline. The White House framed the result as a "commonsense approach" that balanced innovation with security while maintaining American dominance in both AI and cybersecurity.
The order also directed national security agencies to strengthen their cyber defenses through a new "cybersecurity information sharing center." Crucially, it included explicit language stating that nothing in the voluntary disclosure requirement should be interpreted as authorizing the government to impose mandatory licensing, prior authorization, or permits. This carve-out was not accidental. The AI industry had participated directly in shaping the decree, and companies wanted assurance that early review would not become a regulatory stranglehold.
The involvement of major AI firms in drafting the order underscored how closely the government and industry had become intertwined on this issue. OpenAI's global head of public affairs, Chris Lehane, called the order "an important step forward," arguing that safety and innovation must advance together. Microsoft's president, Brad Smith, offered similar praise, welcoming the government's effort to advance innovation while protecting public security. Yet the participation of Anthropic—the company whose Mythos model had sparked the whole initiative—carried additional weight. The Pentagon had previously blacklisted Anthropic and flagged it as a "supply chain risk" after disagreements over safeguards in its models for classified military systems. That the company was now helping to shape government policy on AI oversight illustrated the delicate dance between regulation and industry cooperation.
The path to this order had been neither smooth nor straightforward. Trump had originally planned to sign it on May 20th in a hastily arranged White House ceremony. Executives from major AI and technology companies had been invited, and some were already en route when the signing was abruptly postponed. Trump told reporters that day he had delayed the order because he disliked certain aspects of it and believed it might "obstruct" AI development. He wanted to make sure the decree would not become a barrier to progress. That concern—that government oversight could slow innovation—had apparently resonated enough to send the order back for revision.
The voluntary framework that emerged reflected that tension. Both Anthropic and OpenAI had already begun giving selected companies and governments access to their most advanced models to help them prepare their defenses. Anthropic had announced that Mythos would be released to all its customers within weeks. The National Institute of Standards and Technology had previously announced that major tech companies would share unpublished AI models with the government for national security and public safety evaluations, though that announcement had since been removed from the Commerce Department website. Against this backdrop, Trump's order attempted to formalize and systematize what was already happening informally, while preserving the voluntary nature of the arrangement and explicitly ruling out mandatory government control.
Citações Notáveis
A commonsense approach of collaborating with industry to balance innovation and security while consolidating America's continued global dominance in AI and cybersecurity— White House spokesperson
Safety and innovation must advance together to ensure continued American leadership in AI— Chris Lehane, OpenAI's global head of public affairs
A Conversa do Hearth Outra perspectiva sobre a história
Why did Trump delay the original signing if the goal was to protect American infrastructure?
He worried the initial version was too restrictive—that it would slow down the very innovation he wanted to protect. The thirty-day window in the final version was a compromise between security review and speed.
So the AI companies essentially got what they wanted?
They got a shorter review period than the government initially proposed, yes. But they also agreed to share their most powerful models early. It's a trade: less bureaucratic friction in exchange for transparency.
Why does Anthropic's Mythos model matter so much here?
Because it allegedly can find and exploit cybersecurity weaknesses faster than anything before it. If that's true, the government needs time to patch its defenses before the model becomes widely available.
But the order is voluntary. What happens if a company refuses to share?
The order doesn't say. It's a framework built on cooperation, not enforcement. That's both its strength and its weakness.
Is this really about security, or is it about control?
Probably both. The government genuinely fears what advanced AI could enable in the wrong hands. But it also wants visibility into what's being built. The voluntary structure lets both sides claim they got what they needed.
What comes next?
Watch whether companies actually comply, and whether thirty days is enough time for the government to meaningfully evaluate these systems. If cyberattacks using AI accelerate, the pressure for mandatory oversight will return.