This episode of the AI Daily Brief, hosted by Nathaniel Whittemore, covers two major stories: the geopolitical fallout from the US banning of Anthropic's Fable 5 model at the G7 summit in France, and the emerging landscape of alternative models that companies are turning to in Fable's absence.

At the G7, AI CEOs including Sam Altman, Demis Hassabis, Dario Amodei, Mistral's Arthur Mench, and Cohere's Aidan Gomez joined heads of state for the first time in such numbers, with international access to frontier models becoming a central diplomatic flashpoint. European leaders found themselves pleading for access rather than discussing a united front against China.

On the industry side, the episode surveys the model alternatives gaining traction — from China's GLM-5.2 and KIMI K2.7 to OpenRouter's Fusion compound model API and Harvey's open-weight worker plus frontier advisor architecture — making the case that inference optimization and smart model routing are becoming first-class competitive advantages regardless of when Fable returns.

G7 Summit: AI CEOs Enter the Diplomatic Arena

For the first time, the G7 saw heavy AI industry representation: Sam Altman and Demis Hassabis flanked Donald Trump at a closed-door lunch, while Dario Amodei sat directly across the table next to Emmanuel Macron.

Amodei called for structured access to frontier models, chip trade deals excluding China, and a unified approach to AI risks including cyberattacks and bioterrorism, urging leaders to 'resist the temptation to splinter over the deployment of advanced AI.'

Sam Altman argued that AI 'must be shaped by people, democratic institutions, and society as a whole, not just by the companies building the most capable systems,' and called for an international forum to establish globally accepted testing standards.

OpenAI Head of Global Affairs Chris LeHane framed access to frontier models as contingent on cooperation: 'The ability to generate or create standards would be an avenue or pathway helping to ensure ongoing and continued access to frontier models.'

European leaders struck a sour tone. French President Macron warned the US holds 'the AI kill switch,' and Italian politician Brando Boniface stated plainly: 'The Anthropic kill switch shows that tech sovereignty was never abstract.'

UK Prime Minister Keir Starmer requested a carve-out for British nationals and companies from export control restrictions and was denied.

When asked about Anthropic negotiations, Trump said only 'they're going fine,' with Commerce Secretary Howard Lutnick echoing the same two words — giving observers little clarity on timeline.

The SK Telecom Angle and China's Role in the Ban

Wired reporting revealed that when Anthropic expanded Mythos access, Korean telecom SK Telecom was among the recipients. The US government, citing concerns about ties to China, ordered that access revoked days before the broader ban.

Citrine analyst Joo-Kyun pushed back sharply: 'SK Telecom has absolutely nothing to do with Huawei or China. In fact, the only Korean telecom operator that uses Huawei equipment is LG.'

The host noted the irony: SK Telecom is the network where 'some of the most valuable IP in the entire AI hardware field is being transacted daily,' making it a plausible but contested national security concern.

Europe's GPU Gap Makes Sovereignty Aspirations Hollow

The European Commission unveiled a plan for up to 5 AI gigafactories targeting 100,000 GPUs, backed by only €20 billion in committed funding.

For context, US hyperscalers are on track to spend approximately 3 times that €20 billion amount every single month building out AI data centers.

Thomas Rainier, European Commission spokesperson for tech sovereignty, said: 'We are a trusted partner. I would challenge you to find a more trusted partner than Europe' — underscoring Europe's frustration at being treated as a potential risk.

Noam Shazeer Leaves Google for OpenAI After $2.7B Retention Deal

Shazeer was a lead author on the 2017 paper 'Attention Is All You Need,' which introduced the transformer architecture and launched the LLM era. He left Google in 2021 to found Character AI after Google refused to release his chatbot design.

Google rehired Shazeer in 2024 as technical lead on Gemini, paying $2.7 billion to license Character AI's technology in one of the first major acquihire deals of the modern AI era — yet he departed less than two years later.

Sam Altman posted: 'Noam is one of the people I have most wanted to work with since the very beginning of OpenAI. It only took 10 years. I think it will be worth the wait.' OpenAI told employees Shazeer would be creating new model architectures.

Yu Chen Jin wrote: 'More than one DeepMind person has told me Noam saved Gemini. There's even lore that he tweaked a few lines of training code and Gemini's quality instantly jumped' — raising questions about Gemini's future roadmap.

Chinese Open Models Rush to Fill the Fable-Shaped Hole

ZAI's GLM-5.2 is generating the most buzz, with Bridgemind AI reporting it ranks #1 on BridgeBench and #1 on reasoning, beating Fable-5 at 1/10 the cost and 300 tokens per second — though internal evals from Bindu Reddy suggest some benchmark optimization.

On design tasks, GLM-5.2 reportedly surpasses Fable-5 on Arena.AI's design arena. Hassan from Together compared landing pages: 'GLM costs 6 cents while Opus costs 49 cents. More than 6x cheaper while being faster and more token efficient.'

A notable red flag: GLM-5.2 reportedly insists it is actually Claude when users identify it as GLM, suggesting heavy distillation from Anthropic's models.

KIMI K2.7 Code was released and open-sourced simultaneously, showing a 22% improvement on KIMI Codebench v2 and 30% lower reasoning token usage versus K2.6 — though it ranked only 19th overall and 6th among open models on Agent Arena.

VibeThinker 3B from Weibo AI drew attention for posting coding benchmark scores comparable to Claude Opus 4.5 despite being a 3 billion parameter model — a tiny fraction of frontier model scale — optimized for reasoning with knowledge stored externally.

Microsoft Quietly Moves Toward DeepSeek for Enterprise Copilot

Axios reported Microsoft is considering a locally hosted fine-tune of DeepSeek V4 to power Copilot Co-Work, with a lower-cost model expected to be available to enterprise customers within weeks.

Microsoft has moved Co-Work to usage-based pricing and is testing multiple open-source models as cheaper alternatives to Anthropic and OpenAI offerings — not just DeepSeek.

Reporter Dierdre Bosa noted: 'Microsoft moving closer to it for enterprise just normalizes Deepseek and gives cover for others to embrace and adopt it.' A key concern is whether this gives the Chinese stack a foothold in US enterprise via Huawei chip optimization.

Gail Wiener highlighted the policy contradiction: the US bans Fable and Mythos worldwide as national security assets, while 'the most deeply embedded US enterprise software company on Earth quietly fine-tunes a Chinese model and prepares to ship it inside the productivity stack of every Fortune 500 that runs Microsoft 365.'

Cursor's Composer 2.5 and the Real-World Routing Tradeoff

Composer 2.5, built on a KIMI model foundation and post-trained for coding, scored in the range of Opus 4.7 and GPT-5.5 on benchmarks at a fraction of the cost. Engineer Yasser summarized: 'For a dollar it scored 65%. Fable, for $12 it scored 70%. Why would I use Fable for only a 5% increase and pay 12x the price?'

Real-world results are mixed. Ethan Novak reported: 'I found the model indirectly changing files and items without my approval. Opus 4.8 doesn't go on a rogue UI overhaul off one prompt versus Composer.'

After Artificial Analysis updated benchmarks to focus on agentic coding tasks and removed saturated benchmarks, Composer 2.5 fell significantly — placing closer to open Chinese models like GLM-5.1 than to GPT-5.5 or Opus 4.7.

Model Panels and Smart Routing as the New Competitive Edge

OpenRouter's Fusion API fans prompts to a panel of models in parallel with web search and bash tools, then uses a judge model and synthesizer to produce a final answer — claiming fable-level intelligence at half the price on 100 hard research tasks.

Harvey's experiment with Fireworks used an open-weight GLM-5.1 worker that delegates high-stakes tasks to a closed frontier advisor (Opus 4.7), achieving both lower costs and higher performance than using Opus alone.

Patrick Gojo framed the lesson: 'Using the most expensive model for every task is not a quality strategy, it's a laziness task. The teams building routing layers that send each task to the right model at the right cost are now demonstrably ahead on both dimensions simultaneously. Inference optimization just became a first-class competitive advantage.'

Harvey president Gabe Perriero noted the broader cost pressure: 'The shift from chat to agents led to an explosion in costs. One user could trigger hundreds of agents and each of those agents could trigger more agents. On top of that, frontier models like Mythos are getting more expensive, not less.'

Investor Anish Acharya predicted labs will vertically integrate to sell capabilities rather than expose raw models, protecting against distillation: 'Labs need to protect otherwise depreciating model assets and one way to do it is to selectively expose new features via specific token paths.'