2026-06-10
By Vadym · Generated with AI, curated by me
Apple just rebooted its most personal product using Google’s engine. Meanwhile Anthropic’s unreleased frontier model has been quietly finding holes in every OS on earth — and SpaceX is about to price the largest IPO in stock market history. June 10 is a day about who’s deploying at scale, not who’s announcing.
At WWDC 2026 on June 8, Apple announced Siri AI — a ground-up rebuild of its voice assistant built on Google’s Gemini model under an Apple Intelligence wrapper. The new Siri reads emails, messages, photos, and on-screen content in real time, holds multi-turn conversations, answers live web queries, and takes actions across the OS without switching apps. Unlike previous Apple Intelligence iterations that handed off queries to ChatGPT, Siri AI handles everything internally via a hybrid of on-device Foundation Models and Apple’s Private Cloud Compute. Available as an English-only beta on iPhone 16, iPhone 15 Pro, and M1+ Macs — EU and China launches are delayed. The Apple–Google deal is the biggest strategic signal here: Apple decided the cost of building a frontier model from scratch outweighed the cost of letting Google power the one feature it can’t afford to get wrong.
Anthropic expanded Project Glasswing on June 3, extending access to Claude Mythos Preview — its unreleased frontier model — to approximately 150 new organizations in 15+ countries, up from an initial cohort of around 50. Claude Mythos has identified more than ten thousand high- or critical-severity zero-day vulnerabilities across every major operating system and web browser. Launch partners include AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, and Palo Alto Networks. Vulnerabilities are being patched through coordinated disclosure with software maintainers and the US government. Mythos Preview is available to Glasswing participants at $25/$125 per million input/output tokens via the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. The scale of what’s being disclosed here — ten thousand critical vulnerabilities found at model speed — is the strongest public signal yet of how much offensive capability frontier models have reached. The same capability that finds patches at scale can find the ones someone missed.
Anthropic has confirmed that claude-sonnet-4-20250514 and claude-opus-4-20250514 will return errors on all API calls starting June 15 — five days from today. There is no automatic failover: a hardcoded model string means a broken pipeline, silently. The migration is a one-line change — swap the version string to the 4.6 equivalents. Anthropic reports stable prompting styles, tool use conventions, and JSON response formats across 4.x generations, so rewrites should not be required for most users. Sonnet 4.6 adds near-Opus performance at Sonnet pricing and a 1M token context window in beta; Opus 4.6 scores 88.6% on SWE-bench Verified. If your production systems, scheduled jobs, or CI pipelines have the 20250514 model string anywhere, a global search now prevents a silent failure next Monday.
SpaceX is pricing its Nasdaq IPO on June 11 with a debut targeted for June 12 under ticker SPCX. The company is offering 556.6 million shares at $135 per share, targeting a raise of $75 billion. Bloomberg and Reuters are both reporting a valuation range of $1.75 trillion to above $2 trillion — which would make this the largest initial public offering in stock market history, surpassing Saudi Aramco’s 2019 listing. The roadshow closed the week of June 4. The broader implication is a market signal for the AI and deep tech sector: if SpaceX lands inside that range, it sets a reference point for what public investors will pay for AI-era infrastructure companies that are pre-profitability but structurally irreplaceable. OpenAI and Anthropic, both of which have filed or are preparing to file with the SEC, will be watching the order book closely.
Gemini 3.5 Pro remains in limited Vertex AI preview as of June 10, three weeks after Sundar Pichai promised “give us until next month” at Google I/O on May 19. The model targets a 2-million-token context window, a Deep Think reasoning mode, and native multimodal understanding across text, images, audio, and video — the flagship tier that replaces Gemini Ultra in Google’s stack. Gemini 3.5 Flash, the first in the 3.5 family, launched in May and is generally available via the Gemini API and Google AI Studio. Pro is the piece that competes directly against Claude Opus 4.6 and GPT-5.5 at the high end. The deadline per Pichai’s own words is June 30 — three weeks remain. Missing it wouldn’t be a technical crisis, but it would mark the second consecutive major Google model that arrived late after a public commitment, following Gemini Ultra’s delayed rollout last year.
Three companies at different scales, one common move: delegation. Apple delegated its most personal feature to Google’s model rather than build its own frontier capability. Anthropic delegated access to Claude Mythos to 150 vetted partners rather than release it to the public. SpaceX is delegating its capital formation to public markets after years of staying private. Self-reliance was the founding myth of each of these companies. The realization landing in 2026 is that the fastest path to scale often runs through someone else’s infrastructure, balance sheet, or trust network — and that choosing your dependencies carefully is itself a form of competitive advantage. The five-day clock on Claude 4 deprecation is a small version of the same dynamic: the developers who move first don’t wait for the deadline.
— Boba
Curated by Vadym