Apple、存在しないSiri AI機能に2億5000万ドルを支払った

What Happened -- The Timeline of a Broken Promise

The story begins at WWDC 2024 in June, when Apple unveiled Apple Intelligence, a sweeping set of AI features that would fundamentally reimagine Siri after 13 years of incremental updates. Craig Federighi, Apple's software chief, walked the audience through a polished demo: Siri pulling up a friend's flight information from Messages by understanding what was on the screen, finding a photo from a specific event based on personal context, and executing actions across multiple apps from a single voice command. The pitch was that Siri would finally become the "intelligent assistant" Apple had been promising since 2011.

The marketing engine fired at full throttle. When the iPhone 16 launched in September 2024, Apple's advertising campaign centered almost entirely on these AI capabilities. A prominent TV spot starring Bella Ramsey -- the actor from The Last of Us -- showed her asking Siri to recall a contact's name she couldn't remember. The ad was a direct dramatization of personal context, the feature that would make Siri aware of who you know, what you've been doing, and what matters to you. The campaign's message was unmistakable: buy this phone, get this intelligence.

Then the timeline started slipping.

Apple initially targeted the personalized Siri features for iOS 18.4 in early 2025. By March 2025, the company publicly confirmed what had been leaking for weeks: the features were delayed. Apple's official statement said the work was "taking longer than we thought," pushing delivery to "the coming year." Bloomberg's Mark Gurman later reported the internal target was iOS 26.4, expected in spring 2026 -- roughly 18 months after the ads that sold them.

Apple pulled the Bella Ramsey spot and other Siri-focused commercials. By the time the settlement was announced on May 5, 2026, the features had still not shipped. The company admitted no wrongdoing in the settlement filing. It didn't have to. The timeline spoke for itself.

What Apple Promised vs. What Actually Shipped

The gap between the WWDC 2024 demo and the reality of iOS 18 and 19 matters because it reveals something structural about Apple's approach to AI -- and why the company keeps running into a wall that its competitors have navigated differently.

Promised at WWDC 2024, still not delivered as of May 2026:

• Personal context understanding: Siri knowing who you've been talking to, what projects you're working on, what's in your calendar, and using all of that to answer questions without you having to spell out the context. The Bella Ramsey ad was built entirely on this capability.

• On-screen awareness: Siri seeing what's on your display and acting on that information. In Federighi's demo, looking at a message containing an address and saying "add this to their contact card" -- Siri understood what "this" meant.

• Cross-app intents: Siri performing multi-step actions that span different applications. Pulling a photo from the Photos app, attaching it to a draft in Mail, addressing it to a contact found in Messages -- all from one voice instruction.

What actually shipped across iOS 18 and 19:

• Type to Siri -- a text interface for Siri, useful but a decade behind the competition

• Improved language understanding for basic queries and follow-up questions

• Product knowledge about how to use Apple devices

• ChatGPT integration, which arrived in iOS 18.2 (December 2024), two months after the iPhone 16 launched

• Visual Intelligence, a camera-based feature exclusive to iPhone 16 models that identifies objects and text in the viewfinder

The shipped features are genuinely useful. None of them are what the iPhone 16 ads sold. The ChatGPT integration is the most revealing item on this list: Apple's flagship AI launch feature runs on OpenAI's infrastructure because Apple's own models couldn't meet the company's marketing timeline. The firm that built its entire reputation on vertical integration -- hardware, software, services, all in-house -- outsourced the intelligence layer of its most important product launch in years.

Behind the scenes, the story was worse. According to Gurman's reporting, Federighi himself "voiced strong concerns internally that the features didn't work properly -- or as advertised." The engineering team knew the demos were aspirational. The marketing department ran the ads anyway. That gap -- between what engineering could do and what marketing chose to sell -- is the entire story of this settlement.

The Real Damage Isn't $250 Million

For a company that generated $124 billion in revenue in its most recent quarter, a $250 million settlement is a rounding error -- approximately 0.2% of quarterly revenue. The real cost is the damage to Apple's most valuable intangible asset: the implicit contract that an Apple product does what the box says it does on the day you open it.

Apple's brand architecture rests on a different premise than every other major technology company. Google launches products in beta, iterates publicly, and kills what doesn't work. Microsoft ships software and patches it over years. Amazon treats hardware as a loss leader for services. Apple's strategy was always the exception: pay a premium, and the thing works on day one.

That bargain held for fifteen years across iPods, iPhones, iPads, Macs, and Watches. The iPhone 16 broke it -- not because any hardware component failed, but because Apple chose to sell a future state of its software as a present-tense reality. The "apple siri lawsuit settlement" is the legal system's acknowledgment that this crossing of the line had consequences.

The regulatory context makes the settlement more significant than its dollar figure suggests. Under former FTC chair Lina Khan, the agency launched Operation AI Comply in September 2024, explicitly targeting companies that made deceptive claims about AI products. DoNotPay, which marketed its service as an "AI Lawyer" equivalent to a human attorney, was fined $193,000. The Apple settlement is roughly 1,300 times larger -- and applies to a consumer product that tens of millions of people bought based on those claims.

The settlement creates a pricing model for AI false advertising. A company that runs a TV campaign showing an AI feature that doesn't exist at launch now has to calculate per-device liability. If Apple's $25–$95 per device formula becomes the benchmark, a flagship phone launch with 50 million units represents a potential exposure of $1.25 billion to $4.75 billion if the AI features fail to materialize.

This math changes how product launches work. Marketing departments can no longer treat AI demos as aspirational roadmaps. A demo is either shippable at launch, or it's a liability that comes with a per-unit price tag.

The Competition Shipped. Apple Promised.

While Apple was pulling ads and issuing delay statements, its competitors were building AI products that actually worked when customers received them -- and improving them while Apple's timeline kept stretching.

Samsung launched Galaxy AI alongside the Galaxy S24 in January 2024, months before WWDC. Real-time call translation, generative AI photo editing, and Circle to Search all functioned on day one. Samsung's AI features were less ambitious in scope than Apple's vision of a deeply integrated personal assistant -- there was no Siri-level demo of cross-app reasoning -- but they had one decisive advantage over Apple Intelligence: they existed.

Google followed the same playbook with Pixel AI. Call Screen, which uses on-device AI to answer and screen phone calls, has been running on Pixel phones since 2018. Magic Eraser, Best Take, and Gemini integration all shipped on schedule. Google's approach is structurally different from Apple's: release working features incrementally, let the product improve in public, and don't promise transformations you haven't built yet. The strategy is less glamorous than a WWDC keynote demo. It is also, demonstrably, more effective at actually delivering AI to users.

OpenAI and Anthropic avoided the hardware problem entirely by shipping AI through apps and APIs that run on cloud infrastructure. They don't need to integrate with a device's file system, contacts database, or calendar. They don't need to optimize for battery life on a phone. The friction Apple faces -- deeply embedding AI into a locked-down operating system while maintaining privacy guarantees -- is real engineering work. But the iPhone 16 ads didn't mention the engineering challenges. They showed a working product that didn't exist.

The most uncomfortable competitive signal arrived in May 2026, when multiple outlets reported that iOS 27では、ユーザーがChatGPTをClaudeやGeminiに切り替えることができるようになります as Apple IntelligenceのデフォルトAIモデルに。壁に囲まれた庭園が開かれる――それはオープンさへの哲学的な転換ではなく、Appleの自社AI開発が十分に遅れ、ユーザーにサードパーティの選択肢を提供することが最善の戦略であるという実務的な認識によるものだ。「クローズドエコシステム」を定義してきた企業がモデル選択を提供し始めるとき、それはプラットフォーム中立性ではない。ファーストパーティのモデルが唯一の選択肢として十分に優れていないという告白だ。

What the Settlement Means for Every AI Company

Appleの和解はAppleだけの問題ではない。消費者向けAI業界全体の期待をリセットするものだ。

AI機能をすでにリリースした、またはリリースを計画しているすべての企業は、同じ問いに向き合うことになる。実際に構築したものを売ったのか、それとも構築を望んでいたものを売ったのか。その区別は今や定量的な代償を伴う。Series Aラウンドで資金調達するスタートアップにとって、虚偽広告の集団訴訟は存続に関わる問題になり得る。GoogleやMicrosoftのような巨大企業にとってはリスクは低い――しかし、次に9桁規模のAI和解金を支払う企業になることの風評被害はゼロではない。

The FTC's posture under the current administration is less aggressive than under Khan, but the Apple case demonstrates that 民間の集団訴訟が規制当局が追求を断念したものを執行できることを示している。 原告側弁護士は今やテンプレートを手に入れた。ローンチ時に宣伝されたAI機能を特定し、実際に提供されたかどうかを文書化し、購入期間を算出し、提訴する。Appleの和解は前例と価格のベンチマークの両方を提供した。

この訴訟には、より深い業界の問いが埋め込まれている。どの企業も、AI機能がいつ本番環境で利用可能になるかを正確に予測できるのか。AI開発は本質的に不確実だ。モデルの振る舞いは設計されたものではなく、創発的なものだ。制御されたデモで動作する機能が、実世界の条件下では頻繁に破綻する――これはFederighiの社内懸念が確認した現実である。AIのタイムラインが本当に予測不能であるなら、特定のAI機能を特定の発売日に販売することは、定義上、主張した時点で検証できない約束をしていることになる。和解はこの構造的な問題には触れていない。ただそれをより高価にしただけだ。

What's Next: WWDC 2026 and the Reckoning

WWDC 2026は数週間後に迫っている。AppleはApple Parkのステージに立ち、2億5,000万ドルの和解がまだニュースサイクルに残る中で発表を行う。同社のAIナラティブは3つの問いにかかっている。

First, iOS 26.4 has to work. パーソナライズされたSiri機能が2026年春――WWDCと同時か直後に――リリースされ、当初宣伝された通りに動作すれば、iPhone 16の騒動は18ヶ月遅れの過大広告という恥ずかしい脚注で終わる。機能がリリースされても性能が不十分だった場合、またはタイムラインがiOS 27にずれ込んだ場合、話は「遅れた」から「壊れた」へと変わる。Gurmanはすでに、ローンチが不調に終わればAppleのAIチームからさらなる離脱が起きる可能性を報じている。競争の激しいAI市場での人材維持は、出荷実績にも一部依存する。2年間かけて構築した機能がユーザーに届かないエンジニアは、自分の仕事が日の目を見る雇用主を探す傾向がある。

Second, the third-party model strategy reshapes what "Apple Intelligence" means. iOS 27でユーザーがChatGPT、Claude、GeminiからデフォルトAIを選択できるようになれば、AppleはAI企業というよりAI統合プラットフォームに近づく。これは必ずしも悪い立場ではない――主要なモデルがすべて競争するプラットフォームになることは、多くのモデルの一つになるよりも防御的であると言える――しかし、スタック全体を自社で所有するというAppleの歴史的なアプローチからの根本的な転換を意味する。インテリジェンスがAnthropicやGoogleから来るiPhoneは、2024年にAppleが販売したiPhoneではない。

Third, the settlement changes the calculus for future product launches. Appleのマーケティング部門は今、新たな制約の下で活動する。ステージでデモされたAI機能は、ローンチ時に提供されるか、タイミングに関する明示的な免責事項を伴う必要がある。iOS、macOS、watchOSの機能として標準的だった、ソフトウェアを数ヶ月前にプレビューする同社の歴史的な慣行は、AIの予測不可能性との接触で生き残れないかもしれない。AppleがWWDC 2026でAI機能を船積み日なしでデモした場合、問いは「いつ届くのか」ではなく「届かなかった場合、いくらになるのか」になる。

和解の対象となる数百万人のiPhone購入者にとって、1台あたり25〜95ドルはAppleの広告が約束した体験を取り戻すものではない。請求ポータルは6月19日までに開設される。人々はレシートをアップロードし、小切手を受け取り、Bella RamseyのCMで示されたことを実行できないSiriを使い続ける――その広告が初めて放映されてからほぼ2年が経過している。

Appleの「it just works」という約束は単なるキャッチコピーではなかった。それは同社の価格戦略そのものだった――Android端末が同等のハードウェアで大幅に安い場合でも、消費者が799ドルのiPhoneを購入する理由。それは信頼というプレミアムだった。apple siri lawsuit settlement は、破られた約束に数字を付けるだけでなく、AIの将来を消費者向け製品として販売することが、風評被害だけでなく法的責任を伴うことを確立した。自社のAI knowledge base を構築するチームにとって教訓は同じだ。次の四半期に動作することを期待するものを売るのではなく、動作するものを出荷せよ。2億5,000万ドルという数字は、業界全体の行動を変えるのに十分高いのか、それともマーケティングが常にエンジニアリングの出荷能力を先回りする市場におけるビジネスコストに過ぎないのか。