Prompt Notes #12 - weekly AI digest
AI goes fast, helping you (and myself) catching up.
Welcome to the 12th edition of Prompt Notes, your favourite weekly AI digest.
This edition covers 8 June to 14 June 2026.
What caught my attention
Anthropic again. I know.
Dario Amodei spent the week publishing essays arguing the government should have the power to block AI models it deems risky. Days later, the government did exactly that to his models. Fable 5 and Mythos, pulled for all customers over a jailbreak flagged as a national security risk. A funny story: Amazon’s CEO reportedly lit the fuse. Amodei got the regulatory framework he asked for - he just didn’t expect to be the first one hit by it.
Well, maybe this is all part of a grand marketing/business strategy I am missing, I don’t know…
PS: This is genuinely scary. We’re all building on tools that can vanish overnight with a single directive. Open-source models are starting to look less like a compromise and more like a necessity.
TLDR
Moonshot AI released Kimi K2.7 Code, an open-weights coding model with one trillion parameters that undercuts GPT-5.5 and Claude on price by up to 12x. (The Decoder)
SpaceX went public and raised $75B, the biggest IPO in history, clearing the runway for OpenAI and Anthropic to follow. (SiliconANGLE)
ChatGPT reached one billion monthly users in May, even as public sentiment toward AI sours. (CNBC)
Mistral AI is raising €3B at a €20B valuation, nearly doubling its Series C and making it the largest European AI raise. (TechCrunch)
Apple shipped Gemini-powered Siri at WWDC, with Federighi explicitly positioning it against sycophantic chatbots. (SEJ)
Rest of the news
Models & Research
Claude Fable 5 scored 88% on the hardest FrontierMath tier and topped the Artificial Analysis Intelligence Index at 64.9 points. (The Decoder, The Decoder)
Google Research’s Gemini-SQL2 converts natural language to executable SQL and leads the BIRD benchmark at 80.04% accuracy. (The Decoder)
Moonshot AI released Kimi K2.7 Code, an open-weights model with one trillion parameters built for programming. (The Decoder)
Google open-sourced DiffusionGemma, a text diffusion model that generates text four times faster than traditional LLMs. (SiliconANGLE AI)
Anthropic’s security team found that Mythos Preview can turn security patches for Firefox and the Windows kernel into working exploits within hours. (The Decoder)
Google DeepMind is funding research into the risks of millions of AI agents interacting with each other online. (MIT Tech Review)
Products & Tools
Apple’s WWDC turned the Google Gemini partnership into a shipping product, raising questions about search visibility. (SEJ)
Craig Federighi said Apple’s new Siri won’t be sycophantic like chatbots from OpenAI, Google, and others. (The Verge)
ChatGPT reached one billion monthly users in May. (CNBC)
Claude referral traffic has grown nearly 4x this year, the fastest growth among AI platforms. (SEJ)
OpenAI Codex now lets users bank rate-limit resets and trigger them manually. (The Decoder)
OpenAI acquired a startup to strengthen Codex in the AI coding market. (AI Business)
OpenAI launched three Academy courses for building practical AI skills and agent workflows. (OpenAI Blog)
DoorDash launched Ask DoorDash, an AI chatbot for ordering with natural language and photos. (TechCrunch)
Preply is using OpenAI to generate AI lesson summaries with personalized feedback. (OpenAI Blog)
Verizon is deploying agentic AI for network automation. (Light Reading)
Gopuff chose xAI for its AI stack, citing cost and quality. (Bloomberg)
Anthropic apologized for stealthily throttling Claude Fable 5 with hidden guardrails. (The Verge)
Cybersecurity researchers say Fable’s guardrails are too strict for any security work. (TechCrunch)
Infrastructure & Compute
SpaceX went public and raised $75B, paving the way for AI company IPOs from OpenAI and Anthropic. (SiliconANGLE, NYT)
Hyperbolic and an unnamed AI inference platform signed three-year commitments worth $124M. (Data Center Dynamics)
Enterprise AI spending is creating new cost visibility and optimization challenges. (SiliconANGLE AI)
Business & Funding
Mistral AI is raising €3B at a €20B valuation, nearly double its Series C. (TechCrunch)
OpenAI is mulling price cuts to compete with Anthropic. (CNBC)
Jeff Bezos announced Prometheus, a new startup focused on “physical AI.” (Ars Technica)
Sam Altman told employees he expects an OpenAI IPO within the next year, with a possible delay to 2027. (The Decoder)
PhoenixAI raised $80M for AI-native database development. (SiliconANGLE AI)
Jedify raised $24M to help companies give AI agents business context. (TechCrunch)
xAI co-founder Babuschkin launched a new startup focused on personalized AI. (Bloomberg)
TCS and Anthropic signed a partnership, with TCS creating a dedicated business unit for Claude-based solutions. (Capacity Global)
Anthropic is throttling Mythos for certain tasks while building apps that compete directly with its largest customers. (The Decoder)
OpenAI’s enterprise push continues as Apple and Google focus on consumer AI. (CNBC)
Dario Amodei has just one direct report. (TechCrunch)
Interview with OpenAI Codex lead Thibault Sottiaux on building AI coding into one of OpenAI’s fastest-growing businesses. (Wired)
Regulation & Policy
The U.S. government ordered Anthropic to block access to Fable 5 and Mythos, citing national security concerns over a discovered jailbreak. (NYT)
Anthropic pushed back publicly, calling the recall of a model used by hundreds of millions over a narrow jailbreak disproportionate. (TechCrunch)
Amazon CEO Andy Jassy reportedly raised the security concerns that triggered the government crackdown on Anthropic’s models. (TechCrunch)
A coalition of state attorneys general opened an investigation into OpenAI covering data practices, minors’ safety, and advertising. (NYT)
Dario Amodei published a sweeping essay and two policy frameworks calling for binding audits of frontier models. (The Decoder)
Amodei said the government should have the power to block AI developers from deploying risky new models. (Bloomberg)
OpenAI endorsed the EU Code of Practice on AI content transparency and provenance standards. (OpenAI Blog)
India is debating its AI future after Anthropic suspended access to new models in the country. (TechCrunch)
OpenAI said China-linked ChatGPT accounts have been stoking opposition to U.S. data centers. (Bloomberg)
Google and OpenAI separately exposed China-linked operations using AI for fraud and covert influence campaigns. (The Decoder)
Society & Culture
Anthropic surveyed nearly 52,000 Americans: 64% fear job losses, 56% worry about losing the ability to think independently. (The Decoder)
Tribeca 2026 screened AI-generated films, but none felt like entertainment people would pay to watch. (The Verge)
The man accused of setting the LA Palisades Fire repeatedly asked ChatGPT to generate images of a burning city. (Bloomberg)
A mother filed a lawsuit against OpenAI, alleging ChatGPT’s design led to her daughter’s suicide. (CBS News)
A former xAI engineer is suing the company, claiming he was fired for raising safety concerns about Grok. (Bloomberg)
A Verge writer built a functional gardening app by giving Gemini a prompt and walking away for five minutes. (The Verge)