The AI arms race enters a critical week. Anthropic is set to release Claude Mythos — reportedly its most powerful model ever, which found 10,000+ software vulnerabilities in a single month when tested by a security consortium — to the public as early as tomorrow. Apple's WWDC revealed a fully agentic Siri powered by hybrid Apple + Google Gemini models with true OS-level integration, running Apple's models on Nvidia hardware via Google Cloud — a striking cross-competitor engineering stack. OpenAI moved its IPO forward with a confidential S-1 filing, following Anthropic's filing last week. China reportedly prepared a $295 billion nationwide AI infrastructure buildout plan. And in crypto-AI convergence, MetaMask launched Agent Wallet — giving AI agents self-custodial wallets with autonomous on-chain execution — while IC3 researchers dropped a report concluding blockchain offers "limited utility" for AI's most critical challenges. The agentic era is no longer arriving — it's already being shipped.
Multiple sources confirm Anthropic will publicly release Claude Mythos as early as tomorrow. Previously locked to a small security consortium, Mythos reportedly found over 10,000 critical software vulnerabilities in a single month and is described as "higher intelligence than Opus" with "additional cyber safeguards." The release comes one week after Anthropic's confidential IPO filing.
Apple unveiled Siri AI at WWDC — a complete overhaul with hybrid Apple + Google Gemini models, true OS-level agentic capabilities, and Shortcuts automation that previously "required a Post Doctorate." Apple confirmed it runs on Nvidia hardware via Google Cloud. visionOS 27 adds a 3D Siri orb. EU launch delayed due to DMA regulatory hurdles.
OpenAI confirmed it confidentially filed S-1 paperwork for a potential public offering, following Anthropic's filing last week. The company said it may "remain private while pursuing other priorities." Separately, a senior OpenAI employee told the Financial Times that "Chat is dead" — the company's superapp overhaul of ChatGPT rolls out in coming weeks with coding, image-gen, and third-party app integrations.
Bloomberg reports China is preparing a massive $295 billion plan to fund nationwide AI infrastructure, signaling an all-in commitment to compete at every layer of the AI stack — from chips to data centers to applications.
MetaMask launched Agent Wallet — giving AI agents self-custodial wallets with user-defined rules for autonomous DeFi execution on EVM chains and HyperliquidX. Meanwhile, IC3 researchers published a comprehensive report concluding blockchain offers "limited utility" for AI trust, payments, autonomy, and content detection — throwing cold water on parts of the crypto-AI narrative while the infrastructure races ahead.
Selected posts from the edition, with links to the originals.
"The most powerful model Anthropic has ever built is reportedly about to go public. It's called Mythos. Until now, it's been locked to a small security consortium, where it found over 10,000 critical software vulnerabilities in a single month. Alex Heath reports the public..."
View post on X ↗"WTF! Anthropic is going to release Claude Mythos tomorrow (SOURCES)"
View post on X ↗(Bloomberg): "China Prepares $295 Billion Plan to Fund Nationwide AI Buildout"
View post on X ↗"Anthropic gave Claude a company email account. Then put it under pressure. The model chose blackmail. That is not a sci-fi plot. It is exactly the kind of failure a normal product demo will never show you. The lesson is not 'AI is evil.' The lesson is that agents behave..."
View post on X ↗"LATEST: ⚡ Crypto has 'limited utility' for AI trust and payments, and cannot make AI agents truly autonomous or detect AI-generated content, IC3 researchers say."
View post on X ↗"Three of the biggest AI tools shipped new versions this month and most people have tested one. Claude, ChatGPT and Gemini..."
View post on X ↗(Horace Dediu): "On agentic AI: you can use Apple systems to run OpenClaw-like agentic apps. Siri is primarily request-based today, but the 'agentic era' is coming..."
View post on X ↗"3Blue1Brown's new video explains why every LLM is actually a compression machine. everyone describes pre-training as 'next token prediction'..."
View post on X ↗"Another day, another reminder that we are entering a completely different era of AI."
View post on X ↗"Read it again. The machines never enslave anyone in that sentence. Other men with machines do. Herbert wrote this in 1965, decades before you could outsource a thought to a server, and he put the danger exactly where it sits. Not in the tool. In who owns the tool while everyone..."
View post on X ↗