It's Monday, June 29th: a frontier model gated behind a government review and the largest model-cloning accusation yet both landed in the same week.
Covering what’s happening on the ground in AI, every Monday.
1️⃣ FRONTIER ACCESS: OpenAI Hands GPT-5.6 to Washington First
Image from OpenAI
OpenAI opened a limited preview of GPT-5.6, three new models named Sol, Terra, and Luna, but only about 20 government-vetted partners can use it after the White House asked to review frontier models before public release.
The lineup splits three ways: Sol is the new flagship with a "max" reasoning mode and a sub-agent "ultra" mode at $5/$30 per million tokens, Terra runs $2.50/$15 at roughly half the cost of GPT-5.5 with comparable quality, and Luna lands cheapest at $1/$6.
Access runs through the OpenAI API and Codex only, not ChatGPT, and there is no public waitlist; OpenAI says about 20 vetted partners got in first and expects to widen access in the coming weeks.
The gate traces to Executive Order 14409, signed June 2, which sets up a classified NSA-led benchmarking process and up to 30 days of government pre-release access for any model it deems a "covered frontier model."
Washington's hand isn't new: the Commerce Department forced Anthropic to disable its Mythos and Fable models worldwide this month over fears they'd reach Chinese military users. GPT-5.6 is the second frontier release to ship on the government's terms.
Builders who price around model costs get a real win in Terra and Luna, but only after the model clears a review most teams will never see. Watch whether the 30-day pre-release window OpenAI calls a one-off hardens into the default path every frontier launch has to walk.
2️⃣ DISTILLATION WAR: Anthropic Says Alibaba Ran 28.8 Million Fake Chats to Copy Claude
Image from Pymnts
Anthropic's Head of Policy, Sarah Heck, told US senators that operators tied to Alibaba and its Qwen lab ran roughly 25,000 fake accounts and more than 28.8 million Claude exchanges over six weeks to copy its models into a rival.
In a distillation attack, you flood a leading model with prompts, capture its answers, and use them to train your own, copying its behavior without paying for the research that produced it.
In the letter, Heck said the operators acted "illicitly, systematically, and at industrial scale to harvest U.S. AI capabilities across frontier labs and repackage them as their own without incurring the training and R&D costs."
Anthropic dated the campaign April 22 to June 5 and says the aim was to accelerate China toward its frontier Mythos Preview capabilities.
Anthropic calls it the largest attack of its kind it has seen, topping the 150,000 DeepSeek, 3.4 million Moonshot, and 13 million MiniMax exchanges it flagged in February.
Alibaba, added to the Pentagon's Chinese military companies list this month and challenging that designation, did not respond to a request for comment.
For US labs, the model itself is now the leak, and every API call from a rival is a potential training set. Expect more access controls, identity checks, and rate limits on frontier models, and expect them to land on legitimate developers first.
📰 Other Headlines
CHINA CATCHES UP: China's Zhipu says its open-source GLM-5.2 lands within a point of Claude Opus on a key agentic benchmark at roughly one-fifth the price, gaining enterprise traction as US export limits gate Fable and GPT-5.6.
SPEND DISCIPLINE: Enterprises are ending "tokenmaxxing," with Uber capping AI budgets in monthly tiers after burning a year's spend in four months and Lindy.ai shifting all of its traffic off Claude to DeepSeek.
BETTING ON LLAMA: Internal documents show Meta is building a prediction-market app called "Arena" that uses Llama to auto-generate markets from trending topics and resolve outcomes in near real time.
PHYSICAL AI BET: ON Semiconductor agreed to buy Synaptics for nearly $7 billion, its largest-ever deal, wagering that on-device AI adds $30 billion to its addressable market by 2030.
MEMORY GOES PUBLIC: Shares of SK Hynix, the top supplier of high-bandwidth memory for AI, surged after it filed for a Nasdaq listing seeking up to $29.4 billion.
STATES STEP IN: Rhode Island signed three AI laws covering a therapy-chatbot ban, self-harm safety rules, and AI disclosure in healthcare documentation, as states fill the federal vacuum.
OPEN ON DEVICE: Qualcomm and Hugging Face expanded their partnership to push open, developer-driven AI from edge devices to the cloud.
Your breakdown of what’s happening in AI this week, from Noah Frank ⚡️
🔦 Spotlight On: Exponential View’s View on AI and the Economy
EV releases their v1 report of “The State of the AI Economy” on 6/25.
On June 25, Azeem Azhar's Exponential View, his Substack and research group, claimed to "reconstruct the AI economy from the bottom up." It's a strong piece of work, sixty-odd slides of sharp graphs. If you’d like to read it, you can find the link here: https://substack.com/home/post/p-202866779
As I was reading over the weekend, there were a few graphics that stood out to me. One in particular shows that the firms in the top quartile of AI spend, measured against their own revenue, have grown revenue roughly 92% more than the firms spending nothing, cumulatively since late 2022. Adoption is wildly uneven—I've made that point here often enough—but the size of the spread is new.
To explain the gap, Exponential View leans on Bessen's study of Dutch firms that automated between 2000 and 2016. Those firms grew and hired more, while their incumbent workers lost about 9% of a year's earnings over five years, slowly, concentrated among the older and longer-tenured. That is neither the apocalypse the doom headlines promise nor the free lunch the spend-more crowd implies. But more importantly it may also be the wrong wave to lean on. EV's own deck warns that indexing AI to past cycles understates how fast it moves, and the workers most exposed this time look like the high-skilled professionals earlier automation made more valuable, not the routine ones it displaced. The labor disruption story to date is itself disputed, as Yale Budget Lab still finds no AI footprint or correlation to AI exposure predictions.
None of that argues for doing nothing. It argues for not deciding too much too early. It's tempting, at this stage, to write off whole categories of response—retraining most of all—on the strength of its patchy history. Molly Kinder, one of the sharpest voices on AI and work, calls retraining one tool rather than a fix, worried the frontier will outrun any reskilling. That is a fair caution, and a long way from discarding the tool.
The risk right now is treating the caution as a verdict. If the gap really is about absorption, the levers that we should be protecting are the ones that help workers and firms adapt, and retraining is one of them. The evidence is still thin and mixed, and this is the moment to widen the menu of options, not shrink it.
After all, when it comes to AI and the economy, we must admit what feels scary to a lot of thought leaders: we still don't know what we don't know!
Read the Exponential View report here:
🫵 Want your message in front of 200,000 AI builders?
Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.
For all inquiries, send us a note at [email protected].
The AI Collective is built by volunteers across 180+ chapters in 40 countries.
Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.
🧑💻 About the Editors
About Noah Frank
Noah is a researcher, innovation strategist, and ex-founder thinking and writing about the future of AI and the workforce. His work and body of research explores the economics of emerging technology and organizational strategy. Outside of AIC, Noah heads research for Centaurian AI.
About Joy Dong
Joy is a news editor, writer, and entrepreneur at the intersection of AI and blockchain. Whether she is demystifying complex systems in her newsletter, TEA, or building streamlined solutions through her automation agency, Ownly, Joy’s mission is to make emerging tech accessible and actionable for everyone.
About Lindsay Gross
Lindsay is an AI engineer, researcher, and writer focused on how AI systems behave in practice and what it takes to make them safe. Her work sits at the intersection of AI safety, governance, and product design, and at AIC she writes about the questions that matter most as these systems scale.