Siri Becomes Polygamous and Meta & Co Are Liable for Social Media Addiction
- • Apple opens Siri to ChatGPT and other AI models in iOS 27
- • Meta and Google lose lawsuit and are liable for social media addiction
- • LiteLLM: Fake compliance as a business model
Apple Opens Siri to ChatGPT, Claude, and Gemini Starting with iOS 27
Apple is opening up Siri to competing AI models, ending its exclusive partnership with OpenAI. According to Bloomberg, users will be able to choose between ChatGPT, Anthropic's Claude, and Google's Gemini starting with iOS 27. The integration will be handled through a new “Extensions” feature in Apple Intelligence settings. The shift from an exclusive relationship to an open ecosystem is expected to earn Apple commissions of up to 30 percent via App Store subscriptions. Despite the opening, Google remains deeply embedded in the Apple ecosystem: the search giant is reportedly paying a billion dollars to handle specific tasks within Siri and Apple Intelligence. Apple's own AI development issues led to delays and internal restructuring. → gizmodo.com
Synthszr Take: Apple is monetizing its own weakness. Collecting a 30 percent commission on AI subscriptions while others develop the models—that's platform capitalism in its purest form. Google pays a billion for privileged access, OpenAI gets users instead of cash, and Apple sits in the middle as the gatekeeper. The “Extensions” are not a technical innovation but a tollbooth for foreign intelligence. Cupertino has made a virtue out of necessity: if you can't get your own language models right, you profitably rent out access to your users.
Social Media Addiction: Meta and Google Lose Lawsuit and Are Held Liable
A historic ruling is shaking the tech industry: Meta and YouTube's parent company, Google, have been found guilty of negligence in the first social media addiction lawsuit. The stock market reacted with panic—Meta, Reddit, and Snap lost between 8 and 10 percent, while the Nasdaq only dropped 2.4 percent. The verdict lifts the decades-long liability protection for platforms (Section 230) and opens the door for thousands more lawsuits. Smaller internet companies, in particular, are now under pressure, as any user could theoretically sue over content they have seen. Meta shares are already down 17 percent for the year after the price drop—only Microsoft is performing worse. → Martin Peers
Synthszr Take: Meta knew that millions of children under 13 were on Instagram, and a 2018 strategy document explicitly called for 'getting them in as tweens.' Section 230 no longer protects against design decisions, only against content. 2,000 more lawsuits are waiting. Product teams will now have lawyers in the room when discussing retention features. Infinite scroll and autoplay could soon be regulated like cigarette vending machines: technically possible, but legally and ethically toxic.
Catch Me If You Can: Fake Compliance as a Business Model
A Y-Combinator graduate with 40,000 GitHub stars, 3.4 million daily downloads, and proudly displayed SOC2 and ISO-27001 certificates—LiteLLM had all the hallmarks of a successful open-source project. The tool gives developers access to hundreds of AI models and offers output management, but this week it fell victim to malware injected through an external dependency. The malicious code stole login credentials from everything it touched and set them in motion: more access to open-source packages, more stolen credentials, more proliferation. Researcher Callum McMahon of FutureSearch discovered the malware only because a bug in the sloppily programmed malicious code crashed his computer—even AI luminary Andrej Karpathy suspected 'vibe coding' as the reason for the amateurish implementation. The real scandal: LiteLLM's security certificates came from Delve, another Y Combinator startup accused of generating fake compliance data and collaborating with compliant auditors. → Techcrunch
Synthszr Take: Y Combinator has produced a perfect farce: a hacked flagship startup with fake certificates from another startup that sells fake compliance. 3.4 million daily downloads mean 3.4 million potential backdoors, with the 'SOC2 certified' seal of approval. Supply-chain attacks succeed because no one has a handle on their dependencies. TeamPCP demonstrates the principle perfectly: instead of attacking 100 targets individually, you compromise a central component like Trivy and automatically cash in on everyone who integrates it unversioned. 95 million downloads per month at LiteLLM mean potentially millions of infected systems—and the extracted API-Keys for OpenAI, Anthropic and Co. can be directly monetized. PyPI reacted quickly and quarantined the package, but the damage is done. The industry must finally understand: every unpinned dependency is a ticking time bomb.
Every Launches Hosted OpenClaw Agents for Slack
Every is launching Plus One, hosted OpenClaw agents that can be integrated into Slack with a one-click setup. The startup has spent months developing its own AI employees who take on real responsibilities: R2-C2 handles bug reports and co-writes articles, Iris drafts marketing emails, and Montaigne takes care of growth issues. Every already operates a parallel organizational structure with named AI colleagues, each with a manager and defined areas of responsibility. Hosting is handled via Every-servers, and users can use their ChatGPT-subscription or their own API keys. The rollout is starting with 20 users per week, with Every-subscribers getting preferential access. → Every
Synthszr Take: Every is selling the illusion of a seamless AI colleague while the reality requires Mac Minis and constant supervision. 20 users per week for a hosted service sounds like capacity issues, not deliberate scaling. Plus One solves the infrastructure problem (24/7-machine, token-costs, maintenance), but Every's own experience shows: without dedicated managers and continuous training, AI agents remain expensive toys. The one-click setup promises simplicity where Every itself took months. AI colleagues work, but only with full-time human supervision.
Google Controls Search, but Not Attention
A new analysis of the 5,000 most visited websites shows that marketing influence is distributed across the entire web. Similarweb data proves that search engines account for only 11% of all website visits, while the remaining 89% are distributed across social media, news, e-commerce, and other categories. The study categorized websites into 15 areas and used GPT to classify 4,800 websites. Particularly striking: while 73% of all search queries are made through Google, the actual influence on purchasing decisions has long been elsewhere. The data is based on Similarweb's clickstream-panel, which captures mobile and desktop-browsing behavior worldwide. → TLDR Marketing
Synthszr Take: Google controls search, but not attention. 89% of web visits happen outside of search engines—on social media, news sites, and e-commerce platforms. Marketing teams are pumping billions into Google Ads while their target audiences are long gone. This fragmentation makes performance marketing more expensive and attribution impossible. Anyone still thinking 'Google first' in 2025 is wasting their budget.
AI Agents Bypass the App Store via Direct API Calls
In 2007, Steve Jobs didn't want native apps for the iPhone — only web apps in Safari. A year later, the App Store arrived, combining discovery, distribution, trust, and payments in one controlled layer. Hundreds of billions of dollars in revenue later, this model is on its way out. AI agents don't need graphical interfaces to tap on — they call APIs directly. Anthropic's Model Context Protocol (MCP) became the USB-C for AI integrations in late 2024: a universal standard that OpenAI adopted in March 2025, followed by Google, and which was transferred to a neutral foundation by the end of the year. This breaks down the App Store into three layers: connection (MCP as an open, non-monetized protocol), discovery (the real battleground — who decides if the agent chooses Uber or Lyft?) and payment (almost completely untapped, but potentially the most lucrative as an aggregator for daily microtransactions). → TLDR AI
Synthszr Take: Apple is losing its most valuable asset not through better technology, but through changing interaction patterns. The 30-percent-commission on App Store purchases only works when people have to browse and install apps — agents bypass this intermediate layer completely. MCP as an open standard prevents a single player from monopolizing the connection layer (as Apple did with the App Store). The real battle is happening in discovery: whoever controls the agents' recommendation algorithms will collect the new 200 billion dollars that Google earns from search ads today. Apple's attempt to halt this development with its own AI features falls short — it's optimizing for a world that is already over.
ai;dr is the new tl;dr
A new term is making the rounds and perfectly captures the zeitgeist: AI;DR stands for “AI; didn't read” and is a mutation of the time-honored internet acronym TL;DR (“too long; didn't read”). The semicolon, which originally separated cause and effect – the more you write, the less I read – now separates the machine output from the refusal to pay attention to it. Alberto Romero of The Algorithmic Bridge sees this as a poetic condensation of two civilizational shifts: previously, the length of the text was the barrier to reading; today, it's the suspicion of a lack of human involvement. We have moved from “I'm not finishing this” to “No one started this.” The coming generations are entering the post-literate period of history anyway, while the tech elite is already writing for an audience of AI agents. → The Algorithmic Bridge
Synthszr Take: AI;DR marks the moment when reading culture becomes detective work. Every text is under general suspicion of being machine-generated, even if there's a human behind it. Silicon Valley is responding with WF;AI (“write for the AIs”) and simultaneously optimizing for machines as readers. Post-literacy meets hyper-text production: humans no longer read, machines write more and more. The perfect circle of communicative meaninglessness has closed.
A Nerdy Love Letter to a Format Pronounced Dead
In an interactive essay, Omar Shehata explains how JPEG compression works and why the format, developed in 1992, still processes billions of images daily. The Joint Photographic Experts Group created a standard that removes details invisible to the human eye – maximum visual quality at minimum file size. Shehata's article shows step-by-step how pixels become compressed data: color space conversion, discrete cosine transform, quantization. The kicker: readers can interactively see what happens to the image at each step. Without JPEG, Shehata says, the web would be “less colorful, much slower, and would probably have far fewer cat pictures.” → Parametric Press
Synthszr Take: JPEG survives because it strikes the right balance between physics and psychology. In 1992, the developers understood something fundamental: humans perceive brightness more sharply than color, so the algorithm discards 75% of the color information. Modern formats like WebP or AVIF compress better, but JPEG has critical mass – every device can handle it, and every browser displays it. Tech giants constantly preach 'disruption,' but sometimes the format that is good enough and works everywhere simply wins. Shehata's interactive deconstruction proves it: the best standards are the ones you no longer question.
Why I Love the Constraints of Email
An email veteran with 15 years of experience defends the medium's technical limitations as a creative strength. 600 pixels wide, HTML from 1999, no JavaScript, over 300,000 different rendering variations – what web developers would call broken, the author sees as the ultimate creative challenge. The narrow canvas of a maximum of 700 pixels forces clear decisions: no distractions from parallax scrolling or mega-menus, but pure focus on the core message. Outlook has been rendering emails through the Microsoft Word engine since 2007, Gmail strips style-tags, Dark Mode inverts colors – and it's precisely these limitations that create clarity. When fancy code doesn't work, the content has to carry the weight. → TLDR Marketing
Synthszr Take: Email marketing thrives on the paradox of productive constraint. 600 pixels and table-based HTML force a clarity that modern websites, with their 1440px and infinite possibilities, often lack. Outlook's Word engine as a rendering motor sounds like a technical failure, but it works as a filter against over-design. Marketing teams burn millions on complex web experiences, while the most successful campaigns are often contained in a single-column email. Constraints create focus, and focus sells better than any parallax animation.
