The government takes the controls
The biggest story today isn't a new model — it's who gets to use one. In an unprecedented move, the Trump administration has inserted itself directly into the distribution of the most powerful AI systems, and both leading US labs have complied.
OpenAI launched its most capable model yet, GPT-5.6, on Friday, June 26 — a family headlined by three systems named Sol, Terra and Luna, rolling out gradually across ChatGPT, the API and its Codex tools. But at the request of the US government, most people can't touch it. Reporting aggregated from Reuters, Nextgov/FCW, Broadband Breakfast and others says access is locked to a small group of government-approved partners before any wider release.
Anthropic's story is the mirror image. Its advanced Claude Mythos 5 model was taken offline for roughly 15 days, and the administration is now clearing it to switch back on — but only for vetted hands. According to a Semafor scoop and coverage from WIRED, The Hill, Axios and others, the government has lifted its block and cleared the model for more than 100 US institutions, not the public.
The common thread, surfacing during a cybersecurity context, is striking: Washington now holds an effective veto over who runs frontier AI. For a technology that has been sold as broadly empowering, the new reality is a tightly controlled list of approved organizations.
Anthropic vs. Alibaba: a 44-day cloning campaign
The restrictions land amid an extraordinary security accusation. In a letter, Anthropic has accused Chinese tech giant Alibaba of running what it calls the largest-known "distillation" campaign against its Claude models — a 44-day operation involving 28.8 million queries aimed at copying Claude's most advanced capabilities.
Distillation, in essence, means using a rival's model outputs to train your own, cloning behavior without rebuilding the underlying system. As reported by Reuters, the BBC and others, the scale here is what makes it notable, and it helps explain why the government's caution around frontier releases has sharpened. The fear that cutting-edge models can be siphoned off and replicated is no longer hypothetical.
OpenAI: IPO jitters and an India bet
On the business side, OpenAI is sending mixed signals. The ChatGPT maker may push its planned IPO to 2027, and the mere report of a delay — laid out by its advisers and cited by Livemint via a New York Times report — was enough to send a ripple through global markets. It's a reminder of how much weight investors now place on OpenAI's trajectory.
At the same time, the company is investing in growth. OpenAI has hired longtime Uber executive Prabhjeet Singh as its first Managing Director for India, its fastest-growing market. Singh previously ran Uber's India and South Asia business as President. Building out leadership in India signals that, regulatory gatekeeping aside, OpenAI sees its consumer future in scale.
The bottom line
Today marks a turn. The frontier of AI is no longer defined purely by capability — it's defined by access, and access now runs through Washington. Two of the most powerful models on earth shipped this week behind a government gate, even as one of their makers fends off a massive alleged cloning effort from abroad and the other navigates wobbling IPO expectations. The era of open frontier releases is, at least for the most advanced systems, on pause.