☕️ Trump lifts export ban on Anthropic's most powerful AI

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🔓 Trump lifts export ban on Anthropic's most powerful AI

☁️ Meta to rent out its spare AI computing power

🤖 Anthropic unveils Claude Sonnet 5

🇯🇵 Japan plans sovereign AI and 10 million robots

🎮 Sony will stop making PlayStation game discs in 2028

🍌 Google's new Nano Banana 2 makes AI images in 4 seconds

🎁 + 16 other news you might like

🧰 + 6 trending tools

📚 + 5 trending papers

🔓 Trump lifts export ban on Anthropic's most powerful AI LINK

The U.S. Department of Commerce has withdrawn the export controls it placed on Anthropic's Claude Fable 5 model, letting the company restore global access a day after the ban ended an 18-day standoff.

The fix was a single safety filter, tuned to block one technique Amazon researchers used to prompt Fable 5 into identifying software vulnerabilities, with the new classifier catching that method in more than 99% of cases and rerouting requests to Opus 4.8.

Fable 5 returns across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork today, with AWS, Google Cloud, and Microsoft Foundry to follow, while the June 12th directive had barred all foreign nationals, including Anthropic's non-citizen staff.

☁️ Meta to rent out its spare AI computing power LINK

Meta is looking into starting a cloud infrastructure business that would let customers buy access to its spare AI computing power and its hosted AI models, according to Bloomberg.

The plan, known internally as Meta Compute, would put the company in direct competition with major cloud providers including Amazon Web Services, Google Cloud, and Microsoft Azure.

The effort is reportedly led by infrastructure chief Santosh Janardhan, Meta Superintelligence Labs head Daniel Gross, and company president Dina Powell McCormick, following a similar approach taken by CoreWeave.

🤖 Anthropic unveils Claude Sonnet 5 LINK

Anthropic has released Claude Sonnet 5, which the company describes as a cheaper and faster version of Opus 4.8, letting people run agents at a lower cost even though the version number suggests a bigger jump.

Despite the number 5 in its name, the model is a relatively minor development, and the writer says it will take a few days to properly get a handle on how it performs.

The launch arrives while people still wait for both Claude Fable 5 and GPT-5.6-Sol, and more than 100 American institutions recently regained access to Mythos under a new ad hoc licensing setup.

🇯🇵 Japan plans sovereign AI and 10 million robots LINK

Japan has unveiled a plan to build its own homegrown AI model and put 10 million AI-equipped robots to work across different sectors by the year 2040, led by a consortium called Noetra.

Noetra, which includes SoftBank and Sony, will focus on physical AI, meaning artificial intelligence used in real-world settings like self-driving cars, factory robots, and android butlers, with about $6 billion in government money over five years.

The effort ties into Japan's push to cut its reliance on US and Chinese AI models and address its aging, shrinking population, part of a 14-year plan investing 370 trillion yen across 17 sectors.

🎮 Sony will stop making PlayStation game discs in 2028 LINK

Sony will stop making physical discs for all new PlayStation games in January 2028, shifting to an all-digital model where every new release sells through the PlayStation Store and other retailers.

Games released before the January 2028 cutoff will still be available on disc, and Sony said the change reflects how most players already buy games, with digital downloads making up 85% of full-game software sales on PS4 and PS5.

The decline of physical games shows up in retail too, as GameStop reportedly closed more than 1,300 stores over the past two fiscal years while more players choose to buy their games online.

🍌 Google's new Nano Banana 2 makes AI images in 4 seconds LINK

Google has launched Nano Banana 2 Lite, the fastest and cheapest model in its Nano Banana family, generating AI images in four seconds and costing under four cents per thousand images for developers working at scale.

The model is built for speed rather than quality, aimed at rapid ideation and high-velocity developer pipelines, while keeping reliable prompt adherence, strong character consistency, and legible text rendering inside the images it produces.

Available now in Google AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform, it replaces the original Nano Banana and rolls out to surfaces like AI Mode, the Gemini app, NotebookLM, and Google Photos.

Other news you might like

• Trump earned $1.4 billion from crypto in 2025LINK

• Apple takes Epic fight over app store fees to the Supreme CourtLINK

• Microsoft Plans New Round of Job Cuts as AI and Cloud Spending Tops $100 BillionLINK

• Claude Science, an AI workbench for scientists, is now availableLINK

• Meta will now charge you for the best AI feature on its smart glasses, and there’s a limit even if you payLINK

• Tim Cook and EU tech chief hold ‘constructive’ virtual meeting over Siri AI standoffLINK

• Hidden code in Claude Code secretly flagged Chinese usersLINK

• US House passes online child safety legislationLINK

• Crypto Exchange unveils marketplace where AI agents can hire each otherLINK

• Meta loses bid to dismiss US states' claims that Facebook, Instagram addict childrenLINK

• Amazon is spending billions on deploying engineers into customers looking to get started with AILINK

• Apple Hide My Email Vulnerability Exposes Real Email AddressesLINK

• Google kills Tenor GIF API, forcing changes at X, Discord, and moreLINK

• Realta Fusion generates electricity directly from a fusion reaction, an apparent firstLINK

• Meta considered buying Kalshi before developing its own prediction market appLINK

🧰 Trending tools

Humalike: a behavioral infrastructure toolkit that gives AI agents social skills and proactiveness through APIs, models, and benchmarks.LINK

MailAdept by mailwarm: an AI-assisted service that audits your email infrastructure, fixes deliverability issues, and monitors inbox placement with weekly reporting.LINK

Modelence Mobile Builder: lets you generate full-stack mobile apps from a text prompt, with auth, database, and deployment included out of the box.LINK

Sequence Agentic: lets AI agents programmatically move real money across bank accounts, cards, and loans via one API call with built-in spending limits and audit trails.LINK

Gemini Omni Flash: a multimodal API model that generates and conversationally edits video from text, image, and video inputs at $0.10 per second.LINK

OASIS 1 Ring: a wearable ring combining a trackpad and private voice capture for hands-free text input and editing without a keyboard.LINK

📚 Trending papers & reports

ArXiv, the free science paper library, is planning major changes to how it operates, aiming to better serve researchers and the public as its role in global science keeps growing.LINK

Persuasion by chatbots now beats expert human debaters, trained canvassers, and tournament champions, and was nearly 3x more effective at raising real donations to charity.LINK

Image-generating networks built entirely with transformers now train 4x faster than before, reaching state-of-the-art quality in just 60 epochs by fixing two key scaling problems that previously caused instability.LINK

Fake company data generation now stays internally consistent by separating hard facts from AI-written prose, so teams can safely train and test enterprise tools without corrupting results from contradictory documents.LINK

Secret codes between AI agents can now be hidden inside normal-looking text automatically, meaning standard monitoring tools may no longer catch when two autonomous systems are quietly conspiring.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!