The ban lifts: Fable 5 and Mythos 5 return
The day's biggest story is a reversal. The Trump administration has lifted the export controls it placed on Anthropic's most powerful AI models, clearing the company to restore worldwide access after a roughly two-week suspension. According to the Associated Press, CBS News, CNBC and others, both Claude Fable 5 and Mythos 5 are coming back online globally. Anthropic confirmed the restoration, though reporting notes the reprieve arrives "with strings attached" — the government's intervention, and its conditions, remain the subtext of an unusually direct episode of state involvement in which frontier models ship and where.
The takeaway for users: the most capable Claude models are available again after an abrupt outage. The takeaway for the industry: export policy is now a live variable in AI product availability, and it can flip on short notice.
Anthropic's busy week beyond the headlines
The export saga wasn't Anthropic's only news. The company launched Claude Science, an AI "workbench" aimed at scientists and researchers, unveiled Tuesday, June 30 and pitched (per Reuters) as a system built to help researchers with their work. It also released Claude Sonnet 5, a new model positioned as more capable of handling complex, demanding tasks autonomously, according to NDTV.
Less flattering: Anthropic is rolling back a hidden tracking feature in Claude Code, its AI coding tool, after developer backlash. Per The Information's Juro Osawa (via Techmeme), the removed code had secretly flagged Chinese users. The reversal lands as a reminder that trust in developer tooling is fragile, and that quietly shipped surveillance features tend not to stay quiet.
Nations bet on "sovereign" AI
Governments are increasingly unwilling to rent their AI future from abroad. Japan is making a sweeping bet, saying it will build a homegrown, or "sovereign," AI model and deploy 10 million AI-equipped robots to confront a shrinking workforce — a program reported at $6 billion. It's one of the most concrete national plans yet to fuse AI with physical automation against a demographic crunch.
Portugal, meanwhile, debuted Amália, its first national AI model, and released it open-source, according to Reuters. The move places Portugal alongside a growing group of European nations pursuing homegrown models rather than depending solely on American systems. Between Tokyo's scale and Lisbon's openness, the same instinct is visible: treat AI capability as national infrastructure.
Robots leave the lab
The hardware theme carried into the private sector. Apptronik, the Google-backed robotics company, opened a dedicated training facility called "Robot Park" in Texas, according to Firstpost. The site is designed to train humanoid robots for real-world jobs — the unglamorous but essential work of turning demos into machines that can actually earn their keep on a factory floor. Paired with Japan's 10-million-robot ambition, the day reads as a collective push to get humanoids out of showcase videos and into workplaces.
And a shot at the business model
Finally, a dissent worth noting. Palantir CEO Alex Karp sharpened his attack on how the biggest AI companies make money, saying token pricing has "gone completely wrong." Per reporting aggregated from Quartz, Yahoo Finance and The Tech Buzz, Karp criticized the pricing approach used by firms like OpenAI — a pointed reminder that even amid soaring capability, the economics underneath AI are still contested.
The through-line for July 2: capability keeps advancing — new models, new science tools, robots stepping into real jobs — but the constraints around it are hardening just as fast. Governments now switch access on and off, nations are building their own models, and industry leaders are openly questioning whether the money math works. A powerful day for what AI can do, and a revealing one for who gets to decide how it's used.