Today's Headlines
- Anthropic's Mythos 5 returns to select U.S. critical-infrastructure and defense organizations under government authorization
- SoftBank's Masayoshi Son signals interest in a TEPCO stake — an infrastructure play to attract domestic data centers
- As export limits drag on, Mythos-like models spread across Asia — China's 360 joins in
The center of the day was the partial return of Anthropic's Mythos 5, which had been suspended since June 12 under a U.S. export-control directive. The authorized scope is narrow — select U.S. organizations in critical infrastructure and defense — while a broader reopening and any expansion of access remain in the hands of ongoing talks with the government.
In Japan, the question shifted to where the computation will be powered. SoftBank's Masayoshi Son signaled interest in taking a stake in Tokyo Electric Power (TEPCO) with an eye on attracting data centers to the country. In Asia, meanwhile, newcomers moved into the supply gap that the restrictions had opened. The rule of who may use a model, the power and infrastructure that sustain it, and the competition that fills the vacuum — the state's involvement in AI models moved on a layer separate from raw capability today.
Today's Top Three
Anthropic's Mythos 5 Returns to Select Organizations Under Government Authorization
Anthropic's Mythos 5 model has begun returning to select U.S. organizations in critical infrastructure and defense, under authorization from the U.S. government (follow-up).
Here is the background. On June 12, the U.S. government issued an export-control directive citing national security and asked Anthropic to halt access to Mythos 5 for foreign nationals. Because the company cannot screen users by nationality in real time, it disabled the model for all customers to ensure compliance. That blanket suspension is now lifting, but only in part.
The authorization came on June 26. Anthropic says it is first restoring access for a limited set of critical-infrastructure and defense organizations, and that it will keep working with the government on expanding the scope of Mythos 5 access. A full reopening for the general public — including the top-tier Fable 5 model — has no announced timeline.
For those responsible for adoption, this is more than a recovery notice. A model that was halted for reasons unrelated to its technical merits is now coming back in stages, paced by the progress of talks with the government. The order of who gets to use it first, and for what, remains shaped by policy decisions.
- U.S. government authorizes Anthropic to resume Mythos access for select organizations (ITmedia AI+)
- Claude Mythos 5 now available to select U.S. enterprises; Fable 5 general access still on hold (PC Watch)
- Anthropic's Mythos 5 is back (The Verge)
SoftBank's Son Signals Interest in a TEPCO Stake to Anchor Domestic Data Centers
SoftBank Group's Masayoshi Son has signaled interest in taking a stake in Tokyo Electric Power (TEPCO), Japan's largest electric utility.
The aim is to secure the infrastructure needed to attract data centers to Japan. Because large AI facilities consume enormous amounts of electricity, the idea is to gain visibility over power supply through involvement with a utility, and to pursue that hand in hand with site recruitment.
The move adds a new dimension to industrial policy in the AI era. Discussion in Japan has tended to gather around model performance and use cases; what Son raised is the question one step earlier — how to keep the power and the computing base inside the country. Alongside the rule of who may use a model, the question of who sustains its foundation has surfaced in the same week.
- Son eyes a TEPCO investment in an infrastructure play to attract domestic data centers (ITmedia AI+)
As Export Limits Drag On, Mythos-Like Models Spread Across Asia
With export restrictions on Anthropic's models persisting, models that claim to rival Mythos keep emerging from Asian startups (follow-up).
According to TechCrunch, the latest entrant is China's 360, which unveiled "Tulongfeng," a cybersecurity tool for discovering software vulnerabilities that it says matches Mythos in capability. It follows "Fugu," the agent-orchestration model released earlier by Tokyo-based Sakana AI — a sign that the move to fill the supply gap opened by the restrictions is shifting from a single company's bet to a regional pattern.
These launches coincided with the moment U.S. models became unavailable. Sakana AI co-founder Ren Ito framed the shift as temporary, saying U.S. models remain important to Asia. The longer the restrictions last, however, the more alternatives keep stacking up as real, shipping products across the region.
More From Today
Models and APIs
- OpenAI announced GPT-5.6, which it claims surpasses Fable 5 and Mythos 5. However, at the request of the U.S. government, the company has held back a general release and is offering it only on a limited basis (follow-up). OpenAI unveils GPT-5.6, claiming it beats Fable 5 and Mythos 5 — limited release at U.S. government request (PC Watch)
- Google introduced a new Gemini-powered smart speaker, highlighting its ability to handle multiple commands at once as a step up from earlier models. Google's new Gemini-powered smart speaker handles multiple commands (ASCII.jp)
Products and Features
- An Apple Vision Pro executive is reportedly leaving for OpenAI, continuing the flow of senior product talent from hardware companies to AI firms. Apple Vision Pro exec is reportedly leaving for OpenAI (TechCrunch)
Research
- DeepSeek published a paper proposing "DSpark," a speculative-decoding method to speed up LLM inference, targeting lower latency and cost. DSpark: Speculative decoding accelerates LLM inference (Hacker News)
Policy and Regulation
- Author Margaret Atwood said AI's core problem lies in the quality of its training data, summing it up as "garbage in, garbage out" and adding a cultural voice to the debate over what AI learns from. Margaret Atwood says the problem with AI is 'garbage in, garbage out' (The Verge)
- SoftBank's CEO is among several industry figures questioning the feasibility of Elon Musk's orbital data center concept, as scrutiny of AI-infrastructure expectations grows. SoftBank's CEO isn't the only one with questions about Elon Musk's orbital data center hype (TechCrunch)
Business
- A commentary argues that Big Tech's AI spending is being passed on to consumers through higher prices, asking who really pays for the AI arms race. Why is Apple asking me to pay more for Big Tech's AI obsession? (The Verge)
Source: The editorial team selected these stories from the AI news inbox (collected June 28, 2026 — 18 items, 0 primary and 18 secondary).