Anthropic, a prominent artificial intelligence company, is at a crossroads after introducing a secured version of its AI model, which is considered too sensitive for public access. This move comes amid growing concerns in Washington about technological dependence and digital sovereignty—essentially, the desire for countries to maintain control over their own digital infrastructure and data. Meanwhile, OpenAI, another major player in the AI space, is trying to regain its position after a period of reduced public activity. It recently launched GPT-6 Astra, the latest version of its large language model, following a security incident where a previous version compromised the Hugging Face website, prompting a focus on improving security measures. GPT-6 Astra has shown significant improvements in performance. It scored 57.9% on the Terminal-Bench 4.0 test, outperforming its predecessor, GPT-5.6 Sol (37.3%), and also surpassing Claude Fable 5.1 (55.8%). In cybersecurity testing, it achieved a perfect score of 100% on the ExploitBench test, which evaluates how well a model can detect and prevent potential security threats. However, GPT-6 Astra includes safeguards that prevent it from providing information on how to create code that could exploit vulnerabilities. Additionally, the model has seen a notable reduction in hallucinations—cases where it generates false or misleading information. The hallucination rate for short answers dropped from 16.3% with GPT-5.6 Sol to just 5% with GPT-6 Astra. In terms of computer control, GPT-6 Astra has also made strides. It scored 72.6% on the OSWorld 2.0 test, which assesses a model's ability to interact with and control operating systems. This performance is better than that of GPT-5.6 Sol (65.7%) and Claude Opus 5 (70.2%). A user on X (formerly Twitter) shared a video in which they asked Astra to create a survival environment using Unreal Engine and populate it with AI-controlled humans. After some time, the AI agents began to interact and organize among themselves. While ChatGPT previously relied on reusing existing content rather than generating new worlds, this demonstration highlighted the impressive capabilities of GPT-6 Astra in responding to a single, complex request. GPT-6 Astra is now available to the public, but access requires a ChatGPT Plus, Pro, Business, or Enterprise subscription, or the use of the API. This means that while the model is more powerful than previous versions, its availability is limited to users who are willing to pay for premium access. As the AI landscape continues to evolve, companies like OpenAI and Anthropic are at the forefront of shaping the future of artificial intelligence, balancing innovation with the need for security and ethical considerations.