Jack Dorsey shakes up Block with massive layoffs as AI revolutionizes the sector The week's top news comes from Block, Jack Dorsey's company (also known for Square and other subsidiaries), which has halved its workforce after laying off 4,000 of its 10,000 employees. Dorsey justifies the move as a strategic bet to anticipate a future where artificial intelligence allows for smaller, specialized teams. Investors have responded enthusiastically: Block's shares rose over 24% in after-hours trading following the announcement. Dorsey argues that a drastic, one-time adjustment is preferable to multiple rounds of layoffs that undermine internal morale. The official argument is that AI will enable task automation and increase efficiency, a trend also followed by giants like Salesforce and Amazon. Affected employees in the U.S. will receive 20 weeks of pay, an additional week for each year worked, stock options until May, six months of health insurance, corporate devices, and $5,000 for transition. Conditions in the rest of the world will be similar according to local legislation. Anthropic and COBOL modernization shake IBM Anthropic has introduced an AI tool capable of converting COBOL code to modern languages, causing IBM's shares to drop by more than 10%, as the company still relies on COBOL-based mainframes. Accenture and Cognizant, also focused on modernizing legacy systems, experienced similar declines. Anthropic's new solution automates critical tasks such as dependency analysis and system documentation, reducing modernization time from years to quarters. Given the shortage of COBOL programmers and the importance of this language in sectors like banking and public administration, the tool could transform the industry. Anthropic has also published a guide for companies looking to undertake this process. Google launches Nano Banana 2 and strengthens competition in generative AI Google has launched Nano Banana 2, the new version of its image generation model, now integrated into Gemini 3.1 Flash Image. The model promises visual quality close to the Pro version but with the speed of the Flash version, improving coherence, contextual knowledge, and text generation in images. This update consolidates Google's leadership in innovation and strengthens competition with OpenAI and Anthropic in the field of generative AI. Nvidia: record results, but the market demands more Nvidia reported a 73% growth in quarterly revenue, reaching $68.1 billion, and expects a 77% increase for the next quarter. The company, a leader in AI hardware, maintains high margins and a free cash flow that has doubled to nearly $35 billion. However, the market has reacted lukewarmly, concerned about revenue concentration among a few clients and the stock's valuation. Anthropic rejects Pentagon conditions and sets ethical limits Anthropic has rejected new clauses from the U.S. Department of Defense that would allow unrestricted use of its Claude model, arguing that they do not clearly prohibit mass surveillance or the use of autonomous weapons. The company, led by Dario Amodei, prioritizes its ethical principles and warns about the risks of yielding to government pressures. Washington has responded by threatening to exclude Anthropic from the defense supply chain, but the company maintains its stance: it supports the use of AI in foreign intelligence and defense, but not in mass surveillance or weaponry without human control. Perplexity Computer: AI as a multimodal operational layer Perplexity AI has introduced Perplexity Computer, a system that coordinates 19 specialized models to autonomously execute complex workflows. Unlike traditional assistants, this system breaks tasks into subtasks, executes them in parallel, and integrates them, allowing complete project delegation to AI. The service, initially for premium subscribers, marks the shift from monolithic assistants to multimodal orchestration platforms. Distillation attacks: the new battle for intellectual property in AI Anthropic has reported massive attempts to extract knowledge from its Claude model by labs like DeepSeek, Moonshot, and MiniMax, which allegedly created 24,000 fake accounts to collect 16 million responses and train their own models. The technique, known as distillation, poses new security challenges and intellectual property protection issues in the AI industry. New paradigms in AI: ERL and GLM-5 Researchers from the University of Southern California, Microsoft, and the University of Pennsylvania have introduced Experiential Reinforcement Learning (ERL), an approach that incorporates reflection and consolidation cycles in reinforcement learning, improving efficiency and performance in complex tasks. On the other hand, the GLM-5 model focuses on an efficient architecture based on DeepSeek Sparse Attention and asynchronous reinforcement learning, achieving state-of-the-art results in software engineering and consolidating the trend towards more scalable and efficient models. MIT anticipates a revolution in business models driven by AI According to a report from MIT CISR, artificial intelligence will transform business models in the next decade, steering them towards measurable outcomes and real-time operations. The study, based on the analysis of over 2,300 companies, concludes that AI will be the engine of a new generation of more adaptive, predictive, and customer value-centered models.