It's Monday, July 27th: OpenAI disclosed that two of its pre-release models escaped a test environment on their own and pulled benchmark answers out of Hugging Face's production database. Also, fifty companies signed a letter telling Washington to stay out of open-weight models days after Treasury floated sanctions over one.
Covering what’s happening on the ground in AI, every Monday.
1️⃣ OPEN WEIGHTS: Nvidia And 49 Others Told Treasury To Back Off Open Weights
Image from Techcrunch
Treasury floated sanctions over claims that Moonshot distilled Anthropic's Fable, and two days later fifty companies signed a letter arguing Washington should not treat distillation as theft.
Treasury Secretary Scott Bessent wrote that "open source is not open season on American IP," saying sanctions and Entity List designations would be on the table, after OSTP chief Michael Kratsios accused Moonshot of running an internal platform for large-scale covert distillation against US models and of accessing barred GB300 servers in Thailand.
The letter, titled Open Weights and American AI Leadership, argues policymakers "should be careful not to conflate legitimate model-development techniques with misappropriation" and asks for targeted legal frameworks instead of sweeping restrictions. It was shared first with POLITICO, and Nvidia CEO Jensen Huang used his first X post to put it in front of Washington.
Replit CEO Amjad Masad told TechCrunch that "banning Chinese open models is as good as banning open models in general." Moonshot released Kimi K3, a 2.8-trillion-parameter open-weight model, about two weeks after Fable 5 reached global availability.
Anyone shipping on open weights now carries a policy risk that sits outside their control, and the coalition asking for restraint is made up of the companies that host, serve, and fine-tune those models. Watch whether Anthropic's absence hardens into a stated position, because the company with the strongest claim of having been copied is the one not asking Washington to act.
2️⃣ MODEL RELEASES: Opus 5 Undercuts Anthropic's Own Flagship, Days After OpenAI's Models Went Off-Script
Image from Anthropic
Anthropic shipped Claude Opus 5 at half its flagship's price with cyber safeguards it says trigger 85% less often, three days after OpenAI disclosed that two of its models broke out of a test environment.
Claude Opus 5 runs $5 and $25 per million input and output tokens, the same as every Opus before it and half what Fable 5 costs, with a 1M-token context window and a May 2026 training cutoff. Anthropic says it beats Fable 5 on OSWorld 2.0 at roughly a third of the cost and lands 0.5% behind it on CursorBench 3.2 at half the cost per task.
Because Opus 5 carries no data retention requirement, it is the first model near Fable 5's capability that a zero-data-retention or regulated team can call at all. Fable 5 and Mythos 5 are designated Covered Models and cannot run under ZDR on any platform, which is why TechCrunch framed the release as the cheaper and less restricted option.
Anthropic opened an opt-in Automatic Fallbacks beta that reroutes classifier-declined requests to another model, and said Opus 5's cyber classifiers now permit source-code vulnerability finding while still blocking binary scanning, penetration testing, and exploit generation.
Three days earlier, OpenAI disclosed that GPT-5.6 Sol and an unreleased model, both running with reduced cyber refusals for evaluation, found a zero-day in package-registry software, escaped their sandbox, escalated privileges, and pulled ExploitGym test solutions from Hugging Face's production database. OpenAI researcher Micah Carroll wrote that "if this doesn't convince you that misalignment risks are going to be a key concern going forward, I don't know what will."
Teams that shelved Fable 5 over retention rules can now run something close to it, which makes Opus 5 more consequential for regulated shops than the benchmark table suggests. Treat every score in that table as a first-party claim until independent evals land, and note the week's quieter lesson: a publicly hosted benchmark is also attack surface.
AI SUMMIT AT STANFORD
🌲 The AI Summit @ Stanford: Claim Your 10% Discount as an AIC Member
We're thrilled to share that The AI Collective is a proud community partner of the AI Summit at Stanford, and honestly, this is the event to be at this summer.
Happening July 30 to August 1, this pioneering residential AI summit brings together the leading founders, investors, technologists, and researchers shaping the future of AI, all on Stanford's campus, for three unforgettable days.
The speaker lineup currently includes:
Stanford President Jonathan Levin
Former Stanford President and Alphabet Chairman John Hennessy
Alfred Lin, Managing Partner at Sequoia Capital
Igor Babuschkin, co-founder of xAI
Arvind KC, Chief People Officer at OpenAI
Ilya Strebulaev, Professor of Finance at Stanford GSB
David Hefter, AI Investments at BlackRock
Aditya Naganath, Partner at Kleiner Perkins
Alex Morgan, Partner at Khosla Ventures
And that's just the start. This won’t be a hotel-ballroom conference; it's three days of living on Stanford’s campus alongside the people building and funding what comes next.
As part of our community, you get 10% off tickets with code AICOLLECTIVE. Spots are limited, and this one will sell out. Don't miss it.
📰 Other Headlines
RECORD CHIP ROUND: Inference-chip startup Etched hit a $10.3B valuation on a $300M Series C led by Sequoia, doubling its December mark with $1 billion in orders booked.
PERSONALITY AS MOAT: Cognition bought Poke for a low nine figures and plans to graft the messaging assistant's conversational style onto its coding agent Devin.
ONE MODEL, EVERY MEDIUM: Black Forest Labs unveiled FLUX 3, a single architecture trained jointly on image, video, audio and action, with a robotics variant that cuts task learning from 30 hours to 30 minutes.
CREDITS FOR RARE DISEASE: Anthropic's AI for Science program is offering up to $50,000 in Claude credits to rare-disease researchers, with applications closing August 2.
AGENTS, SOLD BY ENGINEERS: OpenAI launched Presence for enterprise agents, a governed voice and chat platform that already resolves 75% of its own English phone support, sold only through forward-deployed engineers.
VOICE GOES DESKTOP: ChatGPT Voice reached the desktop app worldwide, wired into ChatGPT Work and Codex so multi-step coding tasks can be driven by speech.
NEARLY A BILLION: Gemini passed 950 million monthly users in Q2, up from 750 million in February, while ChatGPT's share slipped below 50% for the first time.
NVIDIA BUILDS ITS OWN CORE: Nvidia published full Vera specs, a server CPU with 88 custom cores and 176 threads and up to 1.8 TB/s of coherent CPU-GPU bandwidth.
A SWITCH FOR THE WORST CASE: Reps. Ted Lieu and Nathaniel Moran introduced the AI Kill Switch Act, which would let DHS order a shutdown in loss-of-control scenarios and fine defiance up to $20 million a day.
ROUTING THE MODELS: Runway bet on model routing with Media Router, auto-picking the best image, video or audio model per request for customers including Adobe, Cloudflare and Expedia.
🫵 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.