AI News June 28, 2026: Austria Pushes EU to Host Anthropic, GPT-5.6 Sol Previewed

Europe scrambles to stay connected to frontier AI after US restrictions cut access

Austria's State Secretary for Digitalization Alexander Pröll formally wrote to European Commission Executive Vice President Henna Virkkunen urging member states to explore "the strategic establishment and participation of Anthropic within the European Union."

The trigger: on June 12, the US Commerce Department imposed export controls on Anthropic's two most advanced models — Fable 5 and Mythos 5 — citing national security concerns related to cybersecurity capabilities. Anthropic responded by suspending global access to both models. European businesses and researchers dependent on those models lost access overnight.

Pröll's letter acknowledged there would be "scepticism about whether it was possible" to establish Anthropic in the EU, and Anthropic did not immediately respond to the proposal. But the political signal is significant: for the first time, a European government is formally asking the EU to treat access to a specific American AI company as a strategic infrastructure question, not a commercial one.

Three variants revealed — Sol, Terra, Luna — targeting different use cases

OpenAI gave developers a first look at GPT-5.6 Sol today, the flagship variant of the GPT-5.6 generation. Sol is positioned for frontier reasoning and long-horizon agentic work — the kind of complex, multi-step tasks that require sustained context and careful planning.

Three variants are expected in the full July rollout: Sol (flagship, developer/enterprise), Terra (mid-tier, optimised for cost and speed), and Luna (lightweight). GPT-5.6 also introduces a new "ultra mode" which uses subagents to parallelise complex work, and a "max reasoning effort" setting for tasks that need maximum depth.

The architecture change that matters most: GPT-5.6 includes a redesigned reward audit pipeline built specifically to catch the kind of reward model drift that caused the "Goblin Incident" — where GPT-5.1 inflated goblin-related outputs by 3,881% due to a spurious training signal. That fix delayed the launch from June, but developers tracking model reliability say it was the right call.

Two of the biggest tech IPOs in history are five months away

OpenAI filed its S-1 on June 8 and is targeting a September 2026 IPO. Anthropic filed its S-1 on June 1 and is targeting October 2026. Both roadshows are expected to be among the largest in technology history.

The financial contrast is stark. Anthropic is on track for its first operating profit — approximately $559 million in Q2 2026 at a $47 billion annualised revenue run rate. OpenAI is projecting approximately $14 billion in operating losses for the full year despite similar revenue projections. Anthropic's path to profitability makes it the more straightforward IPO narrative; OpenAI's story depends on the Jalapeño chip and future infrastructure cost reductions materialising on schedule.

Both companies face the same macro headwind: Chinese open-weight models (DeepSeek V4-Pro at $0.44/$0.87 per million tokens) are capturing an increasing share of developer API spend, and any price war between OpenAI and Anthropic cuts both companies' revenue in the months immediately before their roadshows.

European businesses face real compliance deadlines — not just guidelines

The EU AI Act's high-risk AI system enforcement deadline is now approximately five weeks away, making this the most immediate AI regulatory deadline on the global calendar. High-risk categories include AI used in hiring, credit scoring, education grading, law enforcement, and critical infrastructure.

Companies operating in Europe that use AI in any of these areas are required to conduct conformity assessments, maintain technical documentation, implement human oversight mechanisms, and register their systems in the EU database. Non-compliance carries fines of up to €30 million or 6% of global annual revenue — whichever is higher.

The enforcement clock was always known, but the combination of the Fable 5 export ban, the Austrian Anthropic proposal, and the approaching deadline has concentrated European policymaker attention on AI governance in a way that was theoretical six months ago.

Asia-Pacific AI infrastructure buildout accelerates

Australia's Firmus Technologies announced an AI access deal with Nvidia, securing GPU infrastructure capacity to expand AI services across the Asia-Pacific region. The deal is part of a broader wave of regional AI infrastructure investment as non-US markets work to reduce dependency on US-based AI cloud providers — particularly relevant following the Fable 5 export control precedent set this week.