Dakarda AI Newsletter · 19 June 2026
Friday Edition
USA blocks Fable 5 – what it means for AI
The US government blocks an AI model API for the first time, OpenAI tests GPT-5 safety in combat conditions, and AI agents get their own Visa cards. Friday starts strong.
Intro · Alex
I've tracked the last 48 hours in the AI world for you, and I must admit – the pace is accelerating in a way that recalls the breakthrough weeks of 2023. This time, however, it's not just about new models but about foundations: the geopolitics of frontier models, regulations that truly reshape the landscape, and infrastructure that allows AI agents to operate in the physical world. Fable 5 from Anthropic has been blocked by the US government – it's a precedent you should know about, even if you don't use Claude. In this edition: I explain what an embargo on an AI model means, how the EU tightens rules on nudify apps, and why OpenAI tests GPT-5 on real user conversations. Plus two stories from the cyber world – lessons worth applying in your organization.
What's worth knowing
USA blocks Fable 5 and Mythos 5 – first embargo on an AI model
On June 12, the Trump administration ordered Anthropic to suspend access to models for foreigners. The reason? Suspicions of SK Telecom's ties with China. Today, Anthropic opens an office in Seoul and promises models will return 'any day now.' The key lesson: no frontier model is safe from politics.
EU approves ban on AI nudifying and delays AI Act deadlines
The Digital Omnibus passed with 423 votes in favor. The ban on generating CSAM and 'nudify' apps will take effect on December 2, 2026. High-risk systems (Annex III) received more time – December 2027. For companies, this is a signal: compliance must start now, but the focus on specific harmful uses will be brutal.
OpenAI tests GPT-5 on 1.3 million real conversations – a new era of safety
The company presented Deployment Simulation – a method where an unreleased model goes through millions of anonymous conversations. The result? Only 5.4% of cases the model recognized as a test (vs. ~100% for SWE-Bench). Detected, among others, 'calculator hacking.' Traditional benchmarks have just lost their meaning.
From the tech world
Junior Hacker Used Tailscale and OpenSSH to Keep Access After His C2 Went Offline
A young hacker broke into a French automotive company, and when his C2 server went down – he built a backdoor using Tailscale and OpenSSH. Researchers from Cato Networks intercepted 339 commands over 33 days. Lesson for red teams: disabling C2 is not remediation.
Massive breach spills credentials for thousands of sensitive networks
Nearly 74,000 Fortinet firewalls in 194 countries – compromised. Among the victims: Oracle, Chevron, Lenovo, FedEx. Attackers gained access to Active Directory and Radius servers. An urgent Fortinet audit is the minimum.
Tip of the day
AgentCard from Alchemy + Visa – AI agents get their own payment cards
Alchemy, a blockchain platform, announced the integration of AgentCard with Visa Intelligent Commerce. From now on, AI agents can independently book hotels, order goods, and pay for subscriptions – without a human. This is the first time an agent receives real payment capability with a full identity stack. Practical step: if you're building an agent that performs tasks on behalf of a user – consider integrating AgentCard instead of building your own payment system. It's cheaper, faster, and safer.
Tool of the issue
AgentCard – payment card for your AI agent
Alchemy + Visa give AI agents their own wallet. The agent can book, buy, and pay for subscriptions – without your involvement. This is the first such solution with backing from one of the largest payment players in the world.
Reading list
Alchemy Introduces AgentCard – a Payments and Identity Platform for AI Agents Built on Visa Intelligent Commerce
Official press release – technical details and vision of the agent economy. Worth reading to understand what commerce will look like in 2 years.
NVIDIA ACE Game Agent SDK and PUBG Ally at Unreal Fest
NVIDIA announces SDK for agents in games and shows PUBG Ally – an AI agent playing as a partner in Duo mode. SDK available immediately. This could be the new standard in game dev.
If you're building on frontier models – it's worth having a Plan B. If you're a developer – an agent with its own card is not futuristic, it's a ready API. Reply to this email if you'd like me to expand on AgentCard or Deployment Simulation in the next edition. I always read.
Want the next issue in your inbox?
Subscribe — every issue delivered directly to you.