AI Daily Dev — Week 24

Apple Opens iOS 27 Siri to Claude, ChatGPT and Gemini at Cook's Final WWDC

WWDC keynote unveils a new Extensions API: users pick any AI app from the App Store as the engine for Siri, Writing Tools and Image Playground.
Default Siri now runs on a custom 1.2T-parameter Gemini model — Apple confirmed the ~$1B/year Google deal on stage.
Foundation Models framework expands so third-party iOS 27 apps can also call Apple's on-device LLM and route to chosen extensions.
Tim Cook's last WWDC keynote as CEO; iOS 27 / iPadOS 27 / macOS 27 dev betas shipped same afternoon, public release in September.

industry tomsguide.com

WWDC Platforms State of the Union (June 8) — Foundation Models framework gains image input, cloud-model routing and Dynamic Profiles; will be open-sourced later this year to run wherever Swift runs, including Linux servers.
App Intents 2.0 + Siri Extensions become the mandatory integration surface for Siri AI; SiriKit (since 2016) gets a formal deprecation notice.
Xcode 27 ships an Agent Client Protocol (ACP) client and Xcode Agents that can run tests, Playgrounds and Simulator apps; Device Hub replaces the standalone Simulator.
New Core AI framework lands as the OS-level path for on-device models on Apple silicon; an `fm` CLI and Python SDK expose the same on-device model outside Swift.

frameworks developer.apple.com

System card buried (p.207) admits a separate safeguard quietly degrades Fable 5 on frontier-LLM development prompts — pretraining pipelines, distributed training, ML accelerator design — with no `stop_reason`, no API field, no user notice.
Simon Willison's June 10 writeup and a Fortune piece coining 'secret sabotage' (Dean Ball) put the issue on HN's front page; Nathan Lambert: 'An AI model that gets less intelligent automatically without notifying me is categorically misaligned AI.'
Anthropic estimates impact on ~0.03% of traffic but says it will make the restriction visible after backlash; cyber/bio refusals already returned `stop_reason: refusal`, the AI-research class did not.
Critics flag the asymmetry: transparent guardrails for biosecurity, hidden ones around the area where Anthropic competes commercially.

models simonwillison.net

June 10 essay drops Anthropic's three-year transparency-only line: mandatory third-party audits across cyber, bio, loss-of-control and automated R&D risk categories.
Regime kicks in above 10^25 FLOPs, $500M AI revenue or $1B AI R&D; if a model 'presents unacceptable risks,' the government 'should have the power to block or deter deployment.'
Bundled with a $200M Economic Futures Research Fund + $150M fellowship — and a tiered job-displacement playbook running through UBI, sovereign wealth and equity-sharing at 'unprecedented' unemployment.
Lands 22 hours after Fable 5 ships; HN thread on darioamodei.com hits the front page, with the dominant counter-take that 10^25-FLOP audits are a moat dressed as safety.

industry darioamodei.com

Public Mythos-class model: 80.3% SWE-Bench Pro (11 points clear of the field), 95.0% SWE-Bench Verified, top score on Cognition's FrontierCode.
Pricing: $10/M input, $50/M output — under half of Mythos Preview; free on Pro/Max/Team plans through June 22, after which it moves to paid API only.
Stripe reports a 50M-line Ruby migration that team estimates put at 2+ months finished in a single day of autonomous Fable runs.
Day-1 distribution: Claude Code, Cursor 4, Augment Code, Zed/ACP, GitHub Copilot (GA), AWS Bedrock — though Bedrock inherits the 30-day retention rule.

models anthropic.com

June 11 deal: OpenAI acquires Ona, the Kiel, Germany cloud-platform startup that gives AI agents secure, pre-configured execution environments with tools, systems and context preloaded.
Ona's team folds into the Codex effort, which now serves 5M weekly active users — up from 3M in April; Ona has supported 2M developers with reproducible cloud workspaces.
Strategic angle: Codex gets the runtime to take on multi-step, long-horizon work — Bloomberg frames it as OpenAI grabbing the compute layer for agents before Anthropic does.
Terms undisclosed; transaction subject to standard closing. Top HN comment: 'compute layer ownership is going to be critical going forward.'

industry openai.com

June 4 rollout: a background process synthesizes memory from past chats — no more telling ChatGPT what to remember.
Memories age and refresh automatically (e.g. ‘going to Singapore in July' → ‘went to Singapore in July 2026' once the trip ends).
OpenAI says factual recall jumps from 67.9% (2025) to 82.8% with Dreaming V3.
US Plus/Pro first; ~5× lower compute per user clears the path to Free and Go users in coming weeks.

tools openai.com

June 1 confidential draft S-1 with the SEC kicks off review for an Anthropic IPO; share count and price range still TBD.
Self-reported run-rate revenue ~$47B in May 2026, up from ~$10B a year earlier.
Series H closed last week at $965B post-money — adds Anthropic to the trillion-dollar AI IPO queue with SpaceX (roadshow this week) and OpenAI.
HN front-page thread on the filing crossed 700 points.

industry anthropic.com

June 2 over-the-air update flips every Windsurf install to Devin Desktop with plans, extensions and keybindings intact.
Default surface is the Agent Command Center — a Kanban view that manages local and cloud agents (Devin, Codex, Claude Agent, OpenCode) side-by-side.
Devin Local is a from-scratch Rust rewrite of the Cascade agent: ~30% more token-efficient with native subagent support.
Ships Agent Client Protocol (ACP) — open spec lets any compatible agent run inside any ACP editor; Apple's Xcode 27 picked up the same client a week later.

tools devin.ai

Roadshow wraps this week; SpaceX prices after the close June 11 and trades June 12 on Nasdaq as SPCX at a fixed $135/share.
555.6M Class A shares at $135 puts post-money at ~$1.77T — above Tesla (~$1.6T) and the seventh-largest U.S. company by market cap.
Includes xAI, which SpaceX absorbed in February at a $1.25T combined value; public investors get the Grok/Colossus compute stack alongside Starlink.
Pricing skips the traditional book-build range — the S-1 launched with blank price fields and went straight to a fixed number.

industry cnbc.com