#AI governance

Le dernier modèle d’Anthropic, Mythos, n’est mis à disposition que d’un groupe limité d’organisations, après que des tests internes et externes ont suggéré qu’il peut exploiter des vulnérabilités de manière autonome à un niveau supérieur à celui des modèles précédents

Un nouveau rapport de MIT Technology Review Insights estime que les administrations publiques pourraient mieux être servies par des petits modèles de langage spécialement conçus pour fonctionner sous de fortes contraintes de sécurité, de gouvernance et de connectivité.

OpenAI introduit l’exécution en sandbox pour son Agents SDK, en présentant cette fonctionnalité comme un moyen pour les équipes d’entreprise d’exécuter des workflows automatisés avec un risque mieux maîtrisé lors du passage du prototype à la production.

Scotiabank has launched Scotia Intelligence, an internal AI framework meant to unify data operations, governance, and software tools in a single environment.

A suspect has been arrested after a Molotov cocktail was allegedly thrown at Sam Altman’s San Francisco home, an incident that quickly fed into a wider debate about AI power, rhetoric, and public anxiety.

Anthropic’s decision to keep a powerful new model limited to a small group of organizations highlights how frontier AI deployment is increasingly being shaped by governance, restriction, and staged access.

OpenAI has released a policy blueprint calling for updated laws on AI-generated child sexual abuse material, stronger provider reporting, and safety-by-design requirements for AI systems.

A new debate is emerging around frontier AI firms: the same restraint that can frustrate one government may make a company more attractive to another looking for credibility, control, and policy alignment.

A new enterprise tool is pitching itself as a way to monitor and control autonomous agents as businesses confront a fast-rising shadow AI problem.

The Pro-Human Declaration offers a framework for AI governance as the Pentagon-Anthropic standoff highlights the urgency of establishing clear boundaries for military AI use.

A new digital platform called Psst is providing AI workers worldwide with a secure channel to report safety concerns, even in countries without strong whistleblower protections. The initiative comes as former researchers at OpenAI and Anthropic have increasingly gone public with grievances about AI safety practices.

The Department of Defense is working to align its artificial intelligence providers around a shared set of standards and ethical baselines. The effort comes as the military rapidly expands its use of AI systems while maintaining its commitment to responsible AI principles.