Today's Headlines
- Samsung Electronics rolls out ChatGPT Enterprise and Codex to all Korea staff and its global DX division, reversing a 2023 ban
- Ars Technica publishes an analysis arguing Anthropic's safety messaging may have helped trigger the export ban on Fable and Mythos
- Open-source lab Reflection AI signs a compute deal with SpaceX worth $150 million a month and $6.3 billion in total
The question of how far to embrace AI on the enterprise floor moved sharply today. A company that banned AI three years ago over a data leak is now the one rolling it out across the workforce.
At the same time, the tug-of-war over frontier models continues. As rules tighten on closed models, large-scale compute is flowing toward the open-weight camp, and a new analysis is reopening the question of how regulators and AI companies relate to each other. The frontier of adoption and the frontier of regulation are both in motion.
Today's Top Three
Samsung Electronics Brings ChatGPT Enterprise and Codex Company-Wide
Samsung Electronics is deploying ChatGPT Enterprise and Codex to all employees in Korea and to the global staff of its DX division, which builds Galaxy phones and home appliances.
The rollout reaches both technical and non-technical roles, spanning work from engineering to marketing. Codex is being opened not only for developers writing code but also for non-technical staff building internal tools.
What makes this notable is the reversal it represents. In 2023, Samsung banned generative AI after confidential source code leaked from inside the company into ChatGPT. Three years on, the company has shifted from prohibition to a company-wide rollout.
OpenAI describes this as its largest enterprise deployment to date. Secondary reporting indicates that weekly active use of Codex in Korea grew sharply through the spring, framing the move as a leading example of enterprise AI adoption maturing.
Did Anthropic's Own Safety Rhetoric Help Trigger Its Export Ban?
Ars Technica published an analysis arguing that Anthropic's emphasis on safety in its communications may have helped bring about the export restrictions on its own models.
First, the facts. On June 12, a US government directive suspended access to Anthropic's Fable 5 and Mythos 5 for any foreign national, inside or outside the United States. Ahead of that, researchers had reported a jailbreak that bypassed Mythos 5's cyber guardrails.
From there, the piece turns to argument. It suggests that Anthropic's use of risk-laden language at a higher frequency than its rivals, casting its own models as dangerous, may have shaped how regulators perceived the threat. This is the author's interpretation, not an established cause.
Anthropic itself disputes the ban as overbroad. The company argues that the type of jailbreak cited could apply to other publicly available models as well, and that a narrow finding should not justify cutting access to a commercial model in wide use.
Open-Source Reflection AI Signs a $6.3 Billion Compute Deal with SpaceX
Reflection AI, a lab built around open-source models, has signed a compute deal with SpaceX worth $150 million a month and $6.3 billion in total.
Reflection AI was founded by former Google DeepMind researchers. The deal gives it access to Nvidia GB300 chips at the Colossus 2 data center near Memphis, Tennessee, and runs through 2029.
There is logic on SpaceX's side too. Colossus 2 was originally built for xAI, and SpaceX is monetizing its spare capacity by renting it out, with similar arrangements reported for Anthropic and Google.
The deal is symbolic because of when it lands. As US rules tighten on closed models, the perceived risk of depending on closed systems makes open-weight strategies relatively more valuable. This large contract is read as proof that the open camp can compete at scale.
Other Developments
Models & APIs
- Google made its Interactions API generally available as a single unified interface for Gemini models and agents. It offers server-side state management, background execution, and multimodal generation from one endpoint, and is now the default in Google AI Studio. Interactions API: our primary interface for Gemini models and agents (Google Blog)
- Preferred Networks released PLaMo 3.0 Prime, a from-scratch Japanese large language model. It comes in reasoning and non-reasoning variants and extends context length to 256K tokens. Preferred Networks launches from-scratch Japanese AI PLaMo 3.0 Prime (ITmedia AI+)
- Sakana AI launched Sakana Fugu, a multi-agent platform that auto-routes tasks to the best-suited model. Its top tier, Fugu Ultra, reportedly matches Fable 5 and Mythos Preview by combining specialist models. Sakana AI's Fugu Ultra matches Fable 5 by orchestrating multiple models (PC Watch)
- An industry group founded by Intel and AMD published the full 106-page specification for AI Compute Extensions (ACE), x86 instructions focused on matrix multiplication and reduced-precision formats, advancing from April's conceptual white paper to an implementable reference. 106-page spec published for 'ACE' x86 extensions targeting AI workloads (PC Watch)
Products & Features
- Google DeepMind invested $75 million in film studio A24 to co-develop AI tools for filmmakers, a move alongside Netflix and Amazon that signals AI's deepening reach into Hollywood. Google DeepMind bets $75M on AI's future in Hollywood with A24 deal (TechCrunch)
Policy
- OpenAI unveiled Daybreak, a cyber-defense platform that handles software vulnerabilities from discovery through validation and patching. It also launched Patch the Planet, supporting maintainers of more than 30 open-source projects including cURL, Go, and Python. Daybreak: Tools for securing every organization in the world (OpenAI)
Business
- AI chipmaker Groq confirmed a $650 million raise. It follows Nvidia's roughly $20 billion December 2025 deal that licensed Groq's technology and took its top executives, and Groq is rebuilding with a new leadership team and a pivot to a neocloud business. Groq confirms $650M raise (TechCrunch)
- At Hitachi Investor Day 2026, Chairman and CEO Tokunaga outlined an AI Transformation agenda across energy, mobility, industry, and digital, partnering with Anthropic, Google, Microsoft, and OpenAI. How Hitachi approaches AI transformation: takeaways from CEO Tokunaga (ITmedia AI+)
Other
- Daihatsu deployed an AI visual inspection system at its Ryuoh Plant, developed with VRAIN Solution, to detect tiny defects inside aluminum transmission part bores. Daihatsu automates auto-parts defect inspection with AI (ITmedia AI+)
- Odakyu Electric Railway began live use of an AI system that detects people stranded inside level crossings and automatically triggers an emergency train stop. Odakyu starts live use of AI that detects people trapped at crossings and stops trains (ITmedia AI+)
Source: Selected by the editorial desk from the AI news inbox (collected June 23, 2026; 30 items, 6 primary and 24 secondary).