Skip to content
Daily Brief News Daily Brief

11 AI stories from June 28, 2026: Fable 5 ban nearing its end, Sakana Fugu and China's Tulongfeng fill the vacuum, Pax Silica grows to 24 nations, OpenAI retires GPT-4.5, NO FAKES Act advances, state AI legislation diverges sharply, China bypasses Anthropic geolocks, space data center skeptics multiply, EU enforcement 35 days out, ChatGPT falls below 50% share, and bipartisan AI bill reaches 269 pages

· by Pondero Newsdesk · 11 stories

AI news daily brief: 2026-06-28

Eleven stories today: the Trump administration is days away from restoring Fable 5 access, two Asian AI labs launched cybersecurity-focused models to fill the gap the US export ban created, the Pax Silica semiconductor alliance added 10 nations in two days, and the Senate moved the NO FAKES Act forward unanimously with $750,000 per-violation liability.

Fable 5 ban on track to lift within days as Trump administration nears deal with Anthropic

Axios reported June 27, citing a source close to the situation, that the Trump administration is close to allowing Anthropic to restore public Fable 5 access, which has been offline since June 12. Insiders expect restrictions could be lifted as early as this coming week, per Axios. Commerce Secretary Howard Lutnick's Friday letter confirmed that Anthropic "has worked with the U.S. government to address risks associated with Mythos 5 and Fable 5" and that "these efforts have yielded significant progress," per The Decoder. Mythos 5 was already cleared June 27 for critical infrastructure operators. Operators running fallback models should watch for an official Anthropic announcement expected no later than July 4.

Full story: Fable 5 ban on track to lift within days

Asian AI labs launch Fable rivals as export ban creates a vacuum: Sakana's Fugu and China's Tulongfeng go live

Two Asian AI companies launched models targeting the capability gap left by Anthropic's export-restricted Fable 5, per TechCrunch. Sakana AI (Tokyo, founded by ex-Google researchers) released Fugu Ultra, described as comparable in cybersecurity capability to Fable 5 and aimed at Japanese enterprises and government agencies, per Sakana's release page. China's 360 Security simultaneously unveiled Tulongfeng (Dragon Slayer), a vulnerability-detection model, alongside Yitianzhen, an automated cyber-defense system, per TechRadar. Sakana said the timing was coincidental. 360 made no such claim. Watch for benchmark comparisons once Fable 5 is restored.

Full story: Sakana Fugu and 360 Tulongfeng

Pax Silica alliance grows to 24 members as US pilots AI supply chain credentialing in Panama

The US State Department concluded its Second Pax Silica Summit on June 25-26, 2026. The alliance expanded from 14 to 24 members, adding Germany, the Netherlands, the EU, Argentina, Chile, Costa Rica, El Salvador, Greece, Kazakhstan, and Panama. The State Department announced a pilot AI supply chain credentialing project in Panama designed to fast-track verified semiconductor and critical mineral shipments, per the State Department outcomes release. A Foundry School workforce initiative was announced with Stanford University. The summit was hosted by Under Secretary Jacob Helberg. Watch for whether the Panama pilot generates a replicable model for other member nations.

Full story: Pax Silica grows to 24 nations

OpenAI retires GPT-4.5 from ChatGPT as the last non-5.x model exits the consumer interface

OpenAI retired GPT-4.5 from ChatGPT on June 27, 2026, following a 30-day sunset announced May 28. GPT-4.5 is no longer selectable in ChatGPT or in custom GPTs; the API is unaffected, per OpenAI's model release notes. OpenAI framed the retirement as compute consolidation to free resources for newer models. GPT-4.5 was the last model in the consumer interface predating the GPT-5 series. Paid subscribers who had GPT-4.5 selected were automatically moved to a default GPT-5.x model. o3 is also scheduled to exit ChatGPT on August 26. Watch whether OpenAI announces an API sunset for GPT-4.5 alongside GPT-5.6 Sol's broader rollout.

Full story: GPT-4.5 retirement

Senate Judiciary advances NO FAKES Act unanimously, putting $750,000 liability on unauthorized AI voice and likeness replicas

The Senate Judiciary Committee voted unanimously on June 18, 2026, to advance the NO FAKES Act (S. 4591), sponsored by Senators Blackburn and Coons with 13 bipartisan co-sponsors, per Roll Call. The bill gives every individual a federal intellectual property right over their voice and visual likeness, surviving death by up to 70 years. Platforms that fail to remove unauthorized deepfakes face up to $750,000 in penalties per unauthorized work. The bill now advances to the full Senate. Watch for a floor vote, movement in the House companion bill, and whether platforms announce compliant removal tools ahead of potential passage.

Full story: NO FAKES Act Senate advance

State AI bills surge in 2026: Rhode Island signs three laws, Arizona vetoes all three, California AI teacher ban heads to Newsom

The state AI legislative picture diverged sharply in the past week. Rhode Island Governor Dan McKee signed three AI laws: a chatbot safety bill, a therapy chatbot ban for unlicensed AI services, and a healthcare AI disclosure requirement. Arizona Governor Katie Hobbs vetoed all three AI bills her legislature passed: chatbot safety, synthetic intimate imagery, and AI system adoption, per the Transparency Coalition update. California's AB 2148, which bans AI teachers in public schools, passed both chambers and was sent to Governor Newsom on June 24. New York passed seven AI bills before adjourning June 1, including kids chatbot safety, training data transparency, the FAIR News Act, and a data center moratorium.

Wired investigation: Chinese users are outsmarting Anthropic's Fable 5 geolocation blocks through resellers and VPN networks

Wired published an investigation showing that users in China have routed around Anthropic's geolocation restrictions on Fable 5 and Mythos 5 using VPN services and third-party API resellers that proxy access through US-based accounts, per Wired. The piece highlights the technical challenge of enforcing nationality-based restrictions at the API layer: providers can geofence by IP and billing address but cannot reliably distinguish a Chinese national inside a US company's workspace. The story adds context to why the US government's export-control directive was broader than IP-level blocks alone. Watch whether Anthropic announces enhanced identity-verification measures as part of the Fable 5 restoration terms.

SoftBank CEO and Sam Altman both question Elon Musk's orbital data center pitch

SoftBank CEO Masayoshi Son publicly argued that space-based data centers will not meaningfully reduce AI compute costs and would take too long to develop, echoing skepticism also voiced by Sam Altman, per TechCrunch. Son said "the battle for AI" requires solutions within the next few years, not a decade. Each executive's position aligns with their own business interests: SpaceX benefits from satellite-launch contracts, while SoftBank has committed billions to Earth-based data center investments. Watch whether Musk responds with a concrete hardware or timeline announcement at SpaceX or xAI events.

EU AI Act transparency rules activate August 2, 37 days away: GPAI enforcement and Article 50 compliance deadlines

The EU AI Act's Article 50 transparency obligations activate August 2, 2026, now 35 days out, per the EU AI Act Service Desk timeline. From that date, the EU AI Office's enforcement powers go live: fines, model recalls, and information requests. Any AI system generating synthetic content must disclose that fact to users; GPAI providers must publish technical documentation and comply with copyright transparency requirements. A transitional provision from the May 2026 AI Omnibus gives systems already on the market before August 2 until December 2, 2026, to meet the machine-readable marking requirement under Article 50(2). Minimum action before the deadline: confirm that technical documentation and systemic-risk assessments are filed with the AI Office.

ChatGPT market share drops below 50% for the first time, settling at 46.4% as Gemini climbs to 27.7%

ChatGPT's global AI assistant market share fell to 46.4% by end of May 2026, the first time below 50% since the category emerged, per TechCrunch. Gemini holds 27.7% and Claude holds 10.3%. ChatGPT held 65.3% in December 2024 and has declined steadily since. Despite losing share, ChatGPT crossed one billion monthly users in the same period. The decline is attributed to Gemini's ecosystem integration across Google Search, Workspace, and Android, and to Claude's growing enterprise and developer adoption. Watch whether OpenAI's signaled token price cuts and a broad GPT-5.6 Sol rollout reverse the trend in July.

Great American Artificial Intelligence Act of 2026: bipartisan 269-page discussion draft advances in Congress

Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released a 269-page bipartisan discussion draft of the Great American Artificial Intelligence Act of 2026 on June 4. The bill would impose mandatory transparency reports on large-scale frontier AI developers, require publication of a frontier AI framework, mandate critical-safety incident reporting within 72 hours, and establish whistleblower protections for AI safety employees, per TechPolicy.Press. The draft is circulating for comment and is not yet scheduled for a markup vote. Watch for a markup date announcement and whether the bill's frontier AI definition aligns with the export-control definitions used in the Fable 5 directive.

Sources