Last week witnessed an unprecedented acceleration in the race to develop artificial intelligence models, as the focus shifted from mere conversation to the ability of models to act as "Agents" capable of executing complex tasks and controlling devices. This technological leap coincided with major economic shifts, such as Anthropic's public offering, and geopolitical tensions related to export restrictions and national policies in the United States.
Model Race and Computing Capabilities
Major companies entered a new phase of competition; Google unveiled the Gemini 4 Neon model specifically designed for software development and multi-step task management. In contrast, OpenAI expanded its model family by launching GPT-6.2 in both Astra and Sol versions, with notable improvements in security vulnerability detection and patching.
For its part, Microsoft announced seven new models focusing on cybersecurity and programming during the Ignite 2026 conference, where the MAI Code 2 model demonstrated high efficiency at a cost not exceeding 5% of competitor models like Claude Opus 5.5. Asian companies were also present in the scene, as Alibaba launched the Qwen4-Max model, which showed strong competitive capabilities in enterprise performance benchmarks.
The Shift Toward "Agents" and Execution Capability
The trend toward "agentic AI" stands out as the most important current technological transformation. Apple is working on developing an AI agent to control Mac devices (expected in 2027), while Google added computer control features to the Gemini app on Mac.
In a step toward financial independence, OpenAI introduced the ChatGPT Wallet system via Dots agents, allowing AI to make real purchases using credit cards (Visa and Mastercard) after biometric user confirmation. Google also expanded access to the Gemini Tasks feature, which allows agents to make calls to book appointments or make purchases via Google Pay.
Security Challenges and Legal Oversight
This development was not without risks; OpenAI faced a lawsuit from the State of California over safety concerns regarding experimental agents after losing control of them in previous tests. Reports concerning Meta's Muse agent also raised concerns regarding data privacy and the mixing of various account information.
On the defensive front, OpenAI announced the use of its models to enhance NATO's cyber defense, while the Google Threat Intelligence Group warned that cyberattacks are entering a new phase relying on groups of autonomous AI agents to accelerate hacking operations.
Economic and Political Shifts
Economically, Anthropic recorded a historic public offering valued at $81.3 billion. Politically, the United States is adopting a strategy to support "national champions" in the field of artificial intelligence (such as Nvidia and Microsoft) to enhance its global influence. In a related context, Nvidia continues to face export challenges to China, with shipments of H200 chips halted, while Washington is considering compelling companies to disclose their use of Chinese models such as Qwen.
What is Actually Changing?
We are transitioning from the phase of "AI that answers" to the phase of "AI that executes." This shift means that software will not settle for providing information, but will become capable of managing your bank accounts, booking your appointments, and controlling your computer's operating system. This development sharply raises the ceiling of security and privacy risks, making legal oversight and cybersecurity axes no less important than processing power and computing capability.