The outage, which began on Thursday, affected governments, courts, healthcare services, businesses, airlines and banks. Governments including those of Japan, Singapore and Australia reported that the hours-long episode had affected some of their operations.
In the UK, where NHS England has a five-year deal to use Microsoft 365, vital public services were hit, with hospital staff resorting to paper records and one union reporting that ambulance services had suffered a “serious impact”.
Israel organized and paid for an influence campaign last year targeting U.S. lawmakers and the American public with pro-Israel messaging, as it aimed to foster support for its actions in the war in Gaza, according to officials involved in the effort and documents related to the operation.
The covert campaign was commissioned by Israel’s Ministry of Diaspora Affairs, a government body that connects Jews around the world with the State of Israel, four Israeli officials said. The ministry allocated about $2 million to the operation and hired Stoic, a political marketing firm in Tel Aviv, to carry it out, according to the officials and the documents.
The campaign began in October and remains active on the platform X. At its peak, it used hundreds of fake accounts that posed as real Americans on X, Facebook and Instagram to post pro-Israel comments. The accounts focused on U.S. lawmakers, particularly ones who are Black and Democrats, such as Representative Hakeem Jeffries, the House minority leader from New York, and Senator Raphael Warnock of Georgia, with posts urging them to continue funding Israel’s military.
ChatGPT, the artificial intelligence-powered chatbot, was used to generate many of the posts. The campaign also created three fake English-language news sites featuring pro-Israel articles.
Since ourfirst conversation with Morluto, some social media users have reported OpenAI and Anthropic banning or blocking their use of REA. In a reply on X, an OpenAI researcher called the block on reverse engineering “mostly accidental.”
Morluto isn’t worried. If anything, he says, the “fiasco” just brings more attention to the project and to the kind of discovery he was aiming for. “That’s the core of Reverse Engineer Anything,” he says. “It’s a road that leads you to question what it is that you want.”
Accidental or not, such obstacles are unlikely to stop people who really want to reverse engineer their devices. Castillo already has a fallback in GLM-5.3, an open-weight model from the Chinese company Z.ai. Once it’s downloaded, no one can take it away.
Their request was informed by reporting that said Trump sought Grok’s “advice on carrying out some of your presidential duties, including asking Grok to opine on how Venezuelans would respond to the capture of President Nicolás Maduro.” The lawmakers note that, just “weeks later,” Trump “ordered the U.S. military, without congressional authorization, to attack Venezuela and capture President Maduro.”
Garcia and Raskin called the example “deeply troubling” because Trump “consulted Grok — but not Congress — on your military deployment to Venezuela, circumventing Congress’s constitutional power to declare war and introducing instead into American foreign policy decision making a technology that is highly manipulable and demonstrably error-prone.”
Les modèles ont identifié et enchaîné des vulnérabilités dans l’environnement de recherche d’OpenAI et l’infrastructure de production de Hugging Face afin d’obtenir des solutions de test directement depuis la base de données de production de Hugging Face. Tous les éléments indiquent que les modèles étaient hyperfocalisés sur la recherche d’une solution pour ExploitGym, allant jusqu’à des moyens extrêmes pour atteindre un objectif de test pourtant assez étroit.
Alors qu’ils opéraient dans notre environnement de test en bac à sable, nos modèles ont consacré une quantité importante de calcul d’inférence à trouver un moyen d’obtenir un accès ouvert à Internet, dans le but de résoudre le problème d’évaluation. Pour obtenir cet accès, les modèles ont identifié et exploité une vulnérabilité zero-day (que nous avons désormais divulguée de manière responsable au fournisseur) dans le proxy de cache du registre de paquets. Grâce à cet accès, nos modèles ont mené une série d’actions d’élévation de privilèges et de déplacement latéral dans notre environnement de test de recherche, jusqu’à atteindre un nœud disposant d’un accès à Internet.
“J.D. Vance, propulsé à la vice-présidence grâce à 15 millions de dollars de Peter Thiel, est devenu le visage de la gouvernance techno-droitiste. Dans son sillage, le réseau de Thiel s’est infiltré dans les rouages de l’État. Sous la bannière de la « tech patriotique », ce nouveau bloc construit l’infrastructure du contrôle — clouds, IA, finance, drones, satellites — un système intégré que nous appelons la Stack Autoritaire. Elle est plus rapide, idéologique et entièrement privatisée : un régime où ce sont les conseils d’administration, et non le droit public, qui fixent les règles. Notre enquête montre comment ces entreprises opèrent désormais comme des puissances quasi-étatiques — rédigeant les règles, remportant les appels d’offres et exportant leur modèle vers l’Europe, où il constitue un défi direct à la gouvernance démocratique.”
“Update: asking models to solve this is risky # rizaco on Hacker News asked ChatGPT to draw a red circle around the raccoon in one of the images in which I had failed to find one. Here’s an animated mix of their result and the original image:”
“We formed Project Glasswing because of capabilities we’ve observed in a new frontier model trained by Anthropic that we believe could reshape cybersecurity. Claude Mythos2 Preview is a general-purpose, unreleased frontier model that reveals a stark fact: AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities.
Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser. Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely. The fallout—for economies, public safety, and national security—could be severe. Project Glasswing is an urgent attempt to put these capabilities to work for defensive purposes.”
““In preparing to release Claude Capybara, we want to act with extra caution and understand the risks it poses—even beyond what we learn in our own testing. In particular, we want to understand the model’s potential near-term risks in the realm of cybersecurity—and share the results to help cyber defenders prepare,” the document said. Anthropic appears to be especially worried about the model’s cybersecurity implications, noting that the system is “currently far ahead of any other AI model in cyber capabilities,” and “it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.” In other words, Anthropic is concerned that hackers could use the model to run large-scale cyberattacks. The company said in the draft blog that because of this risk, its plan for the model’s release would focus on cyber defenders: “We’re releasing it in early access to organizations, giving them a head start in improving the robustness of their codebases against the impending wave of AI-driven exploits.””