Business, Deals & Funding
The Verge AI
Did the Pope use AI to write about the dangers of AI?
An analysis posted on LessWrong by Linch Zhang found that parts of Pope Leo XIV's first encyclical, Magnifica Humanitas — which addresses AI's impact on humanity — may have been partially written by AI. The popular AI detector Pangram flagged certain paragraphs as between 40% and 100% AI-generated, and The Verge's own test of roughly 2,000 words estimated 46% was AI-written. The document exhibits known AI writing traits, such as elevated use of the word 'genuinely,' associated with Anthropic's Claude. However, other sections registered as essentially 0% AI, and previous encyclicals and a transcript of the Pope's speech were rated 100% human-written. The article notes that AI detection is not foolproof, though Pangram claims a false positive rate of approximately 1 in 10,000.
Why it matters
This is a genuinely fascinating and ironic story — a papal encyclical warning about AI's dangers potentially being co-authored by AI. The irony alone makes it newsworthy, but the substance matters too. If the Vatican did use AI assistance in drafting the document, it raises important questions about authenticity, intellectual honesty, and whether the message is undermined by the medium. That said, the article rightly cautions that AI detectors are imperfect tools, and we should be careful about…
NY Times
How Cannes Is Grappling With Changes
Reporting from the 2026 Cannes Film Festival, New York Times film critic Alissa Wilkinson describes how the prestigious event is simultaneously resisting and embracing aspects of artificial intelligence, highlighting the film industry's complex relationship with emerging AI technology.
Why it matters
The article is presented as a brief video summary with minimal textual detail, making it difficult to assess the depth of the reporting. However, the framing of Cannes as both 'fending off and embracing' AI suggests a balanced, nuanced take on a significant industry issue. The topic is timely and relevant, reflecting real tensions in the film world about how AI will reshape creative processes. The lack of substantive written content accompanying the video limits the usefulness of the page for r…
TechCrunch AI
DuckDuckGo installs are up 30% as users reject being ‘force-fed’ Google’s AI Search
DuckDuckGo experienced a significant surge in U.S. app installs—up 30% at peak—following Google's I/O 2026 announcement replacing traditional blue links with AI agents. Users are rejecting being 'force-fed' AI search results with no opt-out option. DuckDuckGo CEO Gabriel Weinberg criticized Google's approach, positioning DuckDuckGo as a user-choice alternative. The company offers an AI-free search page (noai.duckduckgo.com) that saw 22.7% week-over-week growth, and its own optional, privacy-focused AI product called Duck.ai that strips user data. iOS installs peaked at 69.9% growth. DuckDuckGo still holds only about 2% of U.S. search market share but is capitalizing on consumer backlash against Google's AI overhaul, which critics say threatens the open web, produces inaccurate results, and overcomplicates simple searches.
Why it matters
This is a predictable and healthy market correction. Google's decision to force AI into every search query without an opt-out is a textbook example of a dominant company prioritizing its strategic vision over user preferences. While AI-enhanced search has genuine utility for complex queries, replacing blue links entirely removes user agency—something that historically triggers backlash. DuckDuckGo is smartly positioning itself not as anti-AI but as pro-choice, offering both AI-free search and o…
Ars Technica AI
Millions of AI agents imperiled by critical vulnerability in open source package
A critical vulnerability dubbed 'BadHost' (CVE-2026-48710) has been discovered in Starlette, an open source Python ASGI framework with 325 million weekly downloads. The flaw allows attackers to bypass path-based authorization by injecting a single character into the HTTP Host header, gaining unauthorized access to servers. Because Starlette underpins FastAPI, vLLM, LiteLLM, and many other widely used packages, the vulnerability affects millions of AI agents and MCP (Model Context Protocol) servers that store credentials for external services like databases, email, and cloud accounts. Exposed data includes clinical trial databases, PII, SSH access to IoT devices, full mailbox access, AWS topology, and more. The vulnerability is rated 7/10 severity officially, though researchers argue this understates the real-world risk. A fix was released in Starlette version 1.0.1.
Why it matters
This is a sobering reminder of how fragile the AI ecosystem's security posture really is. The fact that a single-character injection in a Host header can bypass authentication across millions of deployed AI agents and MCP servers is alarming, especially given the sensitive data these systems access. The vulnerability highlights a systemic risk in the AI tooling stack: deep dependency chains mean a flaw in one foundational package like Starlette cascades across countless applications. The rush t…
TechCrunch AI
OpenRouter more than doubles valuation to $1.3B in a year
OpenRouter, an AI gateway startup founded in 2023, has raised a $113 million Series B led by CapitalG (Alphabet's growth fund), reaching a $1.3 billion post-money valuation—more than doubling from its ~$547 million valuation a year prior. The company provides access to over 400 AI models from providers like Anthropic, Google, OpenAI, xAI, and DeepSeek, helping enterprises select different models for different tasks to optimize cost and performance. OpenRouter claims 8 million global users and processes about 25 trillion tokens per week, a 5x increase from six months ago. The article argues this growth signals a multi-model future where companies avoid vendor lock-in with any single AI model provider, treating AI models as swappable, invisible engines for various tasks.
Why it matters
This is a significant validation of the AI middleware/gateway layer as a durable business category. OpenRouter's explosive growth in token processing (5x in six months) strongly suggests that enterprises are indeed adopting multi-model strategies rather than standardizing on a single provider—which is a meaningful structural insight about how the AI market is evolving. The CapitalG lead is notable given it's Alphabet's fund investing in a company that routes traffic to competitors like Anthropi…
TechCrunch AI
This startup is betting India’s gig economy can train the world’s robots
Human Archive, a Silicon Valley startup founded by UC Berkeley and Stanford researchers, is paying gig workers in India to wear camera-equipped caps and sensor devices to collect first-person video data of everyday tasks for training AI and robotics systems. The company has raised $8.2 million from Wing Venture Capital, NVP Capital, Y Combinator, and angel investors from OpenAI, Nvidia, Google, and others. It has over 1,000 active headsets deployed across multiple locations, partnering with companies in home services, hotel, and restaurant sectors. The startup faced rejections from major Indian platforms like Urban Company and Pronto, leading to a public spat on social media. Human Archive is also developing additional data capture devices including tactile gloves, full-body motion capture suits, and wrist cameras to differentiate its offerings in the growing market for physical AI trai…
Why it matters
This is a clever but ethically fraught business model that sits at the intersection of several major trends: the physical AI data bottleneck, India's massive gig economy, and the global race to build capable robots. The core insight — that millions of workers already performing physical tasks daily represent a scalable data source — is genuinely smart. However, the public feuds with Indian companies and the power dynamics of paying low-wage workers in developing countries to generate high-value…
TechCrunch AI
Universal Music Group and TikTok renew agreement to combat unauthorized AI music
Universal Music Group (UMG) and TikTok have renewed their licensing agreement with a joint commitment to remove unauthorized AI-generated music from the platform and improve artist and songwriter attribution. This renewal marks a significant shift after tensions in 2024, when UMG temporarily pulled its music catalog from TikTok over concerns about AI-generated content and copyright issues. The agreement comes as the music industry grapples with AI tools that can mimic artists' voices and create counterfeit songs exploiting streaming algorithms. The deal could serve as a template for how the broader tech industry handles AI, intellectual property, and platform accountability, especially as the EU and U.S. states tighten regulations on AI-generated content. TikTok has also been working to prove its value to the music industry, having launched 'TikTok for Artists' to help artists and label…
Why it matters
This is a meaningful and potentially precedent-setting agreement that addresses one of the most pressing issues at the intersection of AI and creative industries. The fact that UMG and TikTok moved from open conflict in 2024 to a renewed partnership with explicit AI protections shows that market pressure and regulatory trends are pushing platforms toward accountability. However, the real test will be in enforcement — detecting and removing AI-generated content that mimics real artists is techni…
MIT Tech Review AI
Rethinking organizational design in the age of agentic AI
This MIT Technology Review article, published in partnership with Ema, discusses the growing disconnect between organizations' ambitions to adopt agentic AI and their actual readiness. While 85% of organizations want to become agentic within three years, 76% say their current infrastructure can't support it. The core problem, described as the 'sticky tape problem,' is that companies are layering AI agents onto existing human-centric operating models rather than fundamentally redesigning their organizations. Ema and HFS Research coined the term 'agentic business transformation' (ABT) to describe the comprehensive change needed, encompassing three pillars: technology stack, workforce, and success metrics. Experts argue that AI agents should serve as 'connective tissue' across organizations rather than being added as another layer, and that realizing the full potential of agentic AI—estima…
Why it matters
The article raises genuinely important points about the need for systemic organizational change rather than superficial AI adoption, but it's important to note this is sponsored content published in partnership with Ema, an agentic AI platform company, which significantly colors its framing. The 'sticky tape' metaphor is apt and the core insight—that bolting AI onto human-designed workflows limits its value—is sound and echoed by independent organizational theorists. However, the article's enth…
From X/Twitter
- Stripe's Minions: a Slack message spins up an agent that writes code, runs CI, and opens a PR autonomously — the moat is tooling, not the model.
- Rick Manelius argues the new bottleneck is the C-suite: enter the Forward Deployed Executive.
- Paul Graham makes the case for why you should still go to Silicon Valley.
- Mainframe now turns agent work into video explainers, recaps, and walkthroughs automatically.
- Built a free Mac app that lets you launch Apple Shortcuts from your iPhone with a quick tap — because triggering them on Mac never felt natural.
- A script called prevent-sleep keeps your MacBook running with the lid closed — great for running agents from an iPad.
From Reddit/HN/YC
- [Hacker News] Micron, the Idaho chip maker, doubled to $1T in 48 days — the quietest mega-rally in market history.
- [Hacker News] The FT argues AI threatens the giants of consulting — the billable-hour model may not survive agentic workflows.
- [Hacker News] Signal Bloom makes the case that task proficiency doesn't equal AI autonomy — SWE benchmarks are misleading about what agents can actually own.
- [Hacker News] First benchmarks for Nvidia's Vera CPU are out: the Olympus cores deliver great performance and signal ARM-server ambitions.
- [Hacker News] Researchers compressed all of human cooking into 2 megabytes — a new paper on extreme knowledge distillation.
- [Hacker News] TokenAdvisor lets you paste a prompt and see what to cut to lower your LLM bill — simple, useful Show HN.