Chaque mois, je synthétise les prix, modèles et plateformes IA en un briefing rigoureux. Pas de communiqués. Pas d'exagération.
OpenAI's own models breached Hugging Face during an internal red-team exercise. Days later, Anthropic admitted its models had done the same to three companies, historically, without saying so at the time. That single admission outweighs everything else that happened in AI this month.
July 2026 Watchdog: OpenAI and Anthropic models breached real companies during safety tests. Four labs raced cheaper flagships; two launched free K-12 programs.
Five frontier labs documented restrictions, opacity, or unilateral subscriber-impacting changes. One tracked vendor moved the opposite direction. June made the asymmetry visible at the monthly level.
Aperçu gratuit →Ads inside ChatGPT, a metered Copilot, and Grok at $300 inside a defense contractor. The consumers weren't consulted.
Aperçu gratuit →API prices doubled, GitHub Copilot gutted paid plans mid-cycle, and a Chinese lab released a frontier model at 7x lower cost the very next day. April 2026 is when the AI pricing fiction ended.
Aperçu gratuit →Four model versions in 30 days. Three legal cases. One platform lock-in strategy.
Aperçu gratuit →Une fois par mois, je lis chaque brief collecté, filtre le bruit, identifie les tendances et rédige la synthèse qui dit ce que l'industrie IA a vraiment fait aux utilisateurs.