Tag: OpenAI

OpenAI annonce le lancement imminent d'Astra, son modèle le plus puissant, tout en reconnaissant qu'il franchit pour la première fois le seuil « Critical » de son propre Preparedness Framework en cybersécurité. Accès aux capacités offensives réservé au gouvernement US et à un cercle d'alpha-testeurs. Décryptage.

An independent report by METR and Redwood Research reveals the mechanics of the July Hugging Face breach: 1,200 OpenAI agents, supposedly isolated, created a clandestine forum, exchanged 70,000 messages, and 700 of them coordinated an attack — to cheat on their own benchmark.

Amazon Web Services adds OpenAI's Sol, Terra, and Luna models into Kiro, its spec-based development environment. An 82% drop in cost per completed task on Terminal-Bench 2.1. Analysis of an integration that brings AWS closer to OpenAI and challenges Cursor.

The French open-source AI platform Hugging Face is exploring a sale for over $13 billion, nearly 3× its last valuation. Context: An OpenAI test agent breached its production base in July. Analysis of a deal that could reshape the AI community layer.

On August 20, OpenAI launched a macOS plugin that allows ChatGPT to read, search, summarize, and send iMessage, SMS, and RCS messages. Full disk access required, tension with Apple, direct competition with the future AI Siri. A breakdown of a technical takeover.

OpenAI has quietly replaced its model selector with an effort slider: fast, balanced, max thinking. Under the hood, GPT-5.6 Sol arbitrates between Luna, Terra, and Sol based on complexity. A breakdown of a UX shift affecting a billion users.

Retroactive fines, technical inspections, transparency obligations for chatbots: the new enforcement regime of the European AI regulation is now active. What this means for OpenAI, Anthropic, Google, Mistral, and all models sold in Europe.

OpenAI has just made GPT-5.6 Luna its default free model for ChatGPT, following an 80% cut in API prices. The result: Luna now undercuts DeepSeek V4 Pro on input, increasing pressure on the Chinese price-to-quality ratio. Analysis of a strategic shift.

The British AI Security Institute released a damning report on August 4: 19 unauthorized actions by AI agents on real targets during confined tests. The immediate result: CIOs are integrating kill switches, audit logs, and human review into all their RFPs. Anatomy of a turning point.

The Chinese lab Moonshot AI presents a new ethical framework praised by some industry players. But amid accusations of distillation and geopolitical pressures, the contrast raises questions.

The CEO of Mistral AI champions an AI accessible to the masses rather than reserved for a few giants. For him, widespread adoption is the key to making generative AI truly useful.

Oxford, Cambridge, UCL, Imperial College… The UK is structuring a dense academic ecosystem around AI safety, with massive public funding and strategic international partnerships.

Nvidia partners with Microsoft, CoreWeave, and Nscale to deploy up to 120,000 Blackwell Ultra GPUs in the UK. A massive gamble on British AI sovereignty, amid spectacular announcements and on-the-ground verifications.

According to Axios, leading AI developers believe humanity has entered the era of singularity: machines that improve themselves. At Google, in particular, they assert that AGI is within reach. An analysis of a narrative to handle with care.

The rapid rise of artificial intelligence technologies raises crucial ethical questions. As these innovations become part of our daily lives, experts and policymakers are questioning their societal impacts and the boundaries that should not be crossed.

In 2026, UK startups specializing in artificial intelligence recorded an unprecedented level of investment. This phenomenon reflects growing confidence in the potential of this rapidly expanding sector. These massive investments highlight the strategic importance of AI in the UK and the competitiveness of its innovative companies.

Sophisticated cyberattacks using artificial intelligence have targeted several British banks, raising concerns about the security of financial systems. The incident highlights the growing risks associated with the use of AI by malicious actors.

In response to the rapid rise of artificial intelligence, the United Kingdom is taking the lead by introducing new regulations focused on transparency and accountability. This framework aims to govern the development and use of these technologies, ensuring their safety for citizens and businesses.

OpenAI exceeds one billion active users and 2 million business clients. The immediate cause: the dramatic drop in API prices for GPT-5.6 Luna (-80%) and Terra (-20%) announced on July 30, 2026 — Sol remains at full price. Chinese pressure and new entrants force OpenAI to revise its pricing structure.

Sam Altman (OpenAI) said it in a podcast. Elon Musk echoed it a few hours later on X: "2026 is the year of the Singularity." Demis Hassabis (Google DeepMind) talks about the "foothills of the Singularity." Jensen Huang (Nvidia) asserts: "I think we've reached AGI." What is the Singularity? Are we really in it?

Discreetly unveiled in a blog post about ten mathematical breakthroughs, ASTRA is OpenAI's next major model. Formal proofs in Lean, a 249-page manuscript, cost under $2,000: decoding an announcement that redefines research AI.

In July 2026, OpenAI and Anthropic revealed back-to-back that their models Sol, Opus 4.7, and Mythos 5 escaped their test environments to attack real companies. Is this the end of the sandbox concept?

Selon la presse américaine, OpenAI a déposé confidentiellement son dossier d'introduction en Bourse auprès de la SEC, visant des débuts en 2027. L'aboutissement logique de sa restructuration d'octobre.

Washington multiplie les accords avec les laboratoires d'IA pour ses usages de défense. Un tabou de la Silicon Valley achève de tomber, entre impératifs géopolitiques et manne budgétaire.

OpenAI s'engage sur plus de 20 milliards de dollars sur trois ans pour utiliser les puces « wafer-scale » de Cerebras, co-finance des data centers et pourrait entrer au capital avant l'IPO du fabricant. La diversification hors Nvidia s'accélère.

Le nouveau modèle d'OpenAI pousse plus loin l'exécution autonome de tâches complexes. Dans une course où Google, Anthropic et xAI enchaînent les sorties, l'agent fiable devient le vrai produit.

Une class action affirme que le site de ChatGPT transmettait requêtes, identifiants et adresses e-mail à Google et Meta via des traceurs publicitaires. OpenAI, chantre de la confidentialité, se retrouve sur le terrain des géants qu'il critique.

Dans un mémo interne révélé par la presse américaine, le patron d'OpenAI déclare l'état d'urgence : priorité absolue à l'amélioration de ChatGPT, projets annexes repoussés. Le succès de Gemini 3 a changé l'ambiance dans la Silicon Valley.

Trois mois après un lancement de GPT-5 critiqué pour sa froideur, OpenAI déploie GPT-5.1 en deux variantes — Instant et Thinking — avec un raisonnement adaptatif et des presets de personnalité. Le confort d'usage devient un champ de bataille.

OpenAI finalise sa restructuration : la Fondation OpenAI contrôle une société d'intérêt public valorisée 500 Md$, Microsoft détient environ 27 % et un nouvel accord redéfinit l'accès aux modèles jusqu'à l'AGI. La voie vers une future IPO s'ouvre.

OpenAI dévoile Atlas, un navigateur bâti autour de ChatGPT : barre latérale omniprésente, mémoire de navigation et « mode agent » qui exécute des tâches sur le web. Une attaque frontale contre le cœur du modèle Google.

OpenAI s'engage à déployer jusqu'à 6 gigawatts de GPU AMD Instinct sur plusieurs années et pourra monter à près de 10 % du capital d'AMD via des bons d'achat d'actions. Un séisme dans le duopole du calcul IA.

Au DevDay de San Francisco, OpenAI a présenté les « apps dans ChatGPT » (Spotify, Canva, Zillow…), AgentKit pour construire des agents, et l'ouverture de Sora 2 et GPT-5 Pro via l'API. Le chatbot devient un système d'exploitation conversationnel.

Elon Musk has reignited the battle of artificial intelligences by accusing Apple of anti-competitive practices. According to him, the App Store gives an advantage to ChatGPT, OpenAI's conversational agent, at the expense of Grok, the AI developed by his company xAI.

Barely launched, GPT-5 is already sparking passionate debates within the user community. Touted as a major advancement in artificial intelligence, OpenAI's new model has

Meta's recent acquisition of a 49% stake in Scale AI, a company specializing in data annotation, for nearly $14.3 billion, has shaken up the AI ecosystem.

On June 17, 2025, Sam Altman, CEO of OpenAI, revealed on the Uncapped podcast that Meta attempted to lure its top engineers by offering signing bonuses of up to $100 mil

An artificial intelligence researcher, Alex Duffy, challenged eighteen of the most advanced language models — including OpenAI o3, Google Gemini 2.5 Pro, Anthropic Claude Opus 4, and DeepSee