Today's Headlines

  • SoftBank Group commits another ¥1.63 trillion to OpenAI, with the third tranche due October 1
  • Cloudflare pushes AI companies to pay publishers for their content, setting a September 15 deadline
  • Anthropic redeploys Claude Fable 5 as export controls are lifted and global access resumes
  • SoftBank subsidiary SB Intuitions releases the Japanese-language "Sarashina3" LLM series
  • Meta prepares "Meta Compute" to sell off its spare AI infrastructure capacity

The lead story today was not a smarter model. It was the money and the rules that keep AI running.

Investment, content fees, and export policy all surfaced on the same day. Each turns on a question that sits beneath the performance race: who bankrolls the AI business, on what terms, and who bears the cost. Away from the benchmark headlines, the financial and regulatory scaffolding for putting AI into society is being rebuilt.

Today's Top Three

SoftBank Group commits another ¥1.63 trillion to OpenAI

SoftBank Group announced it will inject another $10 billion (about ¥1.63 trillion) into OpenAI on October 1, the third tranche of its staged investment plan.

The commitment builds on earlier payments. By SoftBank's account, the first tranche of $30 billion was announced on February 27 and executed on April 1, and a second tranche of $10 billion was completed in March. With this third tranche, the total under the plan is set to reach roughly $64.6 billion (about ¥10.3 trillion).

That figure is separate from the company's existing stake, which has accumulated to about $34.6 billion since September 2024. Delivered in successive tranches, the funding steadily deepens the ties between SoftBank and OpenAI.

Dated July 1, the announcement is a reminder that a foundation-model company can keep drawing sums on the scale of a national megaproject, quietly, alongside the noisier competition over model quality.

Cloudflare presses AI companies to pay publishers for their content

Cloudflare unveiled a policy that requires AI companies to separate the crawlers they use for traditional search from those used for AI training and agent work.

Firms that fail to make that separation by September 15, 2026 will, by default, be blocked from crawling pages that host ads on Cloudflare-protected sites. The rule applies to new customers, to new sites added by existing customers, and to all free-tier users. In effect, "mixed-use" crawlers are shut out of the ad-supported pages that fund publishers.

Alongside the rule, Cloudflare set up a "Pay Per Use" mechanism so publishers can be compensated when their content is used. An evolution of the company's earlier "Pay Per Crawl" marketplace, it triggers payment not simply when content is fetched, but when that content creates value—for instance by appearing in search results or when premium material is accessed. Ceramic.ai and You.com are named among the early partners.

CEO Matthew Prince has noted that most traffic on the internet is already non-human. The move reworks the assumption that AI can vacuum up the web's information for free, steering value back toward the publishers who produce it.

Anthropic redeploys Claude Fable 5 as the export ban lifts (follow-up)

Anthropic redeployed its top model, Claude Fable 5, on July 1, restoring access around the world after an export-control suspension.

This closes a story we have followed for weeks. On June 12, the U.S. Commerce Department applied export controls to Claude Fable 5 and Mythos 5 and cut off access for all users. The trigger was a jailbreak technique that could bypass safeguards to surface software vulnerabilities; Anthropic argued the flaw was narrow, and the matter escalated into a standoff with the administration. The controls were lifted on June 30, leading to the July 1 redeployment.

The models return with tighter safety measures. Anthropic added a classifier aimed squarely at the reported jailbreak, which it says blocks the technique in over 99% of cases. Requests that trip the safeguards are routed to the higher-end Opus 4.8, and the company widened its safety margin to catch more ambiguous cybersecurity requests. As a result, some routine work such as coding will temporarily fall back to Opus 4.8.

Availability differs by model. Claude Fable 5 is offered broadly with safeguards, while Mythos 5 is limited to U.S. organizations. On the paid Pro, Max, Team, and select Enterprise plans, Fable 5 is available at up to 50% of weekly usage limits through July 7. With that, a safety dispute that had grown into a political flashpoint reached a resting point.

Other Developments

Models

Products

Regulation & Policy

  • Nintendo president Shuntaro Furukawa aired the company's stance on generative AI. He framed the technology positively, noting game development and AI have long been linked, while flagging intellectual-property protection and power consumption as concerns and stressing that Nintendo will act firmly against IP infringement regardless of whether generative AI is involved. Nintendo airs its stance on generative AI (ITmedia AI+)

Business

Source: Selected by our editorial team from the AI news inbox collected on July 2, 2026 (30 items; 5 primary, 25 secondary).