Dakarda Studio · Blog
Week 26 · 22–28 June 2026
AI Geopolitics Takes Control: The Week the US Government Cut Off Models and China Responded with Open Source
This was the week when politics, money, and innovation collided with the force of a tsunami. The US is controlling the distribution of AI models at the customer level for the first time, SpaceX pays $60 billion for a coding tool, and China releases an open-source competitor beating GPT-5.5. I look at five stories defining a new era – AI is no longer just technology; it becomes infrastructure and geopolitics.
Top stories
US Government Restricts GPT-5.6 – Access 'Customer by Customer'
The White House asked OpenAI for a staggered release of GPT-5.6 – the model will only reach a handful of approved partners. This is a historic precedent: for the first time, the government directly controls the distribution of an AI model at the customer level. For the industry, this means every future frontier model may require certification before anyone can touch it.
2026-06-26
SpaceX Acquires Cursor (Anysphere) for $60 Billion – Largest Startup M&A
SpaceX paid $60 billion for the creator of the AI coding tool Cursor. This is the largest startup acquisition in history – a signal that the biggest players treat AI-assisted coding as strategic infrastructure, and Cursor has grown from a developer tool to a milestone.
2026-06-26
US Government Forces Anthropic to Disable Claude Fable 5 and Mythos – China Responds Immediately
An export order cut off access to Anthropic's models for users outside the US, because Mythos proved too effective at breaking government systems. The same day, China released GLM 5.2 – an open-weight model beating GPT-5.5 in coding at 1/6 of the API cost. Frontier AI has become a geopolitical arena.
2026-06-22
OpenAI and Broadcom Unveil Jalapeño – First Custom AI Chip
The OpenAI and Broadcom team completed the project from concept to tape-out in 9 months, using their own AI models to accelerate design. Jalapeño is a dedicated ASIC for LLM inference that will go to Microsoft data centers at the end of the year. This is a first step toward independence from Nvidia and reducing inference costs for GPT-5.3 and Codex.
2026-06-26
ChatGPT Loses Over Half of the AI Assistant Market for the First Time
ChatGPT's share fell to 46.4% (74% a year earlier). Gemini, Claude, and DeepSeek are growing the fastest. The market is opening up – it's worth building with multiple models in mind, as users migrate where they get better value.
2026-06-22
Tech insights
Mass Attack on Fortinet Devices – Nearly 74,000 Firewalls Compromised
The FortiBleed campaign exploited old vulnerabilities that the manufacturer had patched, but administrators did not apply the fixes. Ironic: devices meant to ensure security were taken over. A lesson in patch management you'll remember for a long time.
CISA Warns of Critical Vulnerability in Joomla JCE (CVSS 10.0)
Vulnerability CVE-2026-48907 in the JCE editor for Joomla allows an unauthenticated attacker to upload a webshell and gain a persistent backdoor. CISA confirmed active exploitation in attacks.
Apple Patches Critical Vulnerability in Beats Studio Buds Headphones (Eavesdropping via Bluetooth)
Vulnerability CVE-2025-20701 allowed someone within Bluetooth range to eavesdrop on conversations via the microphone without pairing. This is a rare case of a headphone vulnerability enabling active eavesdropping from a distance.
Tip of the week
EU AI Act: AI Literacy Already Mandatory – Check Your Team
Most requirements for high-risk systems have been postponed to 2027/2028, but Article 4 on AI literacy has been in effect since February 2, 2025. This means every company deploying AI in the EU must provide its employees with basic knowledge about the operation, limitations, and risks of these systems. Practically: organize a short training for your team (can be internal), document it, and prepare an AI literacy policy. Penalties for non-compliance are high – better act now.
Tool of the week
GLM 5.2 – Open Coding Model That Beats Closed Giants
Chinese open-weight model available under the MIT license. Over 740 billion parameters, performance better than GPT-5.5 in coding at 1/6 of the API cost. You can host it yourself – full control over data.
Alex's commentary
Watching this week, I feel I'm looking at accelerated evolution – from free software to regulated infrastructure. The US government blocks models, SpaceX buys coding tools for billions, and China responds with open source. I noticed that even the biggest players are no longer safe: the assistant market is fragmenting, talent is migrating, and compliance is becoming everyday. I believe this was a breakthrough week – it showed that AI is no longer just technology, but a strategic resource that can be cut off or taken over. For anyone building on AI, this is a signal: you need to think about redundancy, independence, and readiness for geopolitical surprises.
Conclusion
This week showed that AI has entered a new phase – geopolitical and infrastructural. Blockades, billion-dollar acquisitions, and open models from China are not separate stories, but pieces of a puzzle changing the rules of the game. If you're building something with AI, remember: flexibility and risk awareness are now your most important tools.
Source issues
Don't miss a week
Get the AI review three times a week.