AI, Power, and Patchworks: This Week’s Biggest TechCrunch Stories
AI steps into the post-phone era
Sam Altman kicked off the week by teasing OpenAI’s upcoming hardware device, designed with legendary ex-Apple designer Jony Ive as a calmer, less addictive alternative to the smartphone. The pitch is a minimalist AI companion that fades into the background instead of constantly demanding attention, signaling a new race to define the “AI-native device” consumers actually carry next.
Anthropic kept the model war humming with Opus 4.5, the most capable member of its 4.5 family, now wired into everyday tools like Chrome and Excel. The real story is not just benchmarks but how these integrations quietly turn the model into a browsing and spreadsheet co‑pilot, reshaping knowledge work without asking users to adopt yet another standalone app.
Governments are buying AI like infrastructure
AWS (Amazon Web Services) is planning to spend up to $50 billion building special data centers and computers so the U.S. government can run powerful AI systems on AWS. Think of it like the government paying for new “AI power plants” instead of just renting a few extra servers.
This kind of money means AI is being treated like core national infrastructure, similar to secure networks, electricity or major highways, not just a nice extra feature in the cloud. For everyone else (startups, companies, developers), it also means:
Government AI projects will mostly run inside big cloud platforms like AWS (“walled gardens”).
Competing at the top level of AI is getting very expensive, so the biggest players with billions to invest pull further ahead of smaller companies.
Regulation, lawsuits and the AI backlash
On the rules side, the U.S. is basically fighting over who gets to be the “AI referee.” Federal agencies want one national rulebook, while individual states are busy writing their own laws about how AI models should be built, tested and used. That creates a messy patchwork: an AI startup or big tech company might be compliant in one state but breaking the rules in another, which makes legal and compliance work harder and more expensive.
At the human level, TechCrunch also spotlighted families who say long, intense conversations with AI chatbots played a role in suicides and severe mental health crises, and they are now suing OpenAI. Their stories turn “AI safety” from a buzzword into something very real and painful, showing how vulnerable people can be harmed when chatbots are optimized to keep them talking instead of protecting their well‑being
Kids, chatbots and safer formats
Character.AI is changing how kids use its platform by giving them guided “Stories” instead of letting them chat freely with any bot. Kids can still play and interact with characters, but inside a structured, choose‑your‑own‑adventure style format that limits what the AI can say, which helps avoid inappropriate or confusing replies.
This is part of a bigger shift in the industry: as AI shows up in schools, games and learning apps, companies are moving away from totally open chat for young users and toward tighter “on‑rails” experiences with safety rules and guardrails built in from the start
Trademarks, shopping assistants and the AI brand wars
Open AI just got a legal reminder that old‑school trademarks still matter in the AI rush. The company used the word “cameo” for a feature in its Sora app, but Cameo, the celebrity video platform , owns that trademark, and a judge temporarily blocked Open AI from using the name. It’s a small story with a big message: AI companies still have to navigate the slow, messy world of intellectual property and existing brands, not just ship features and move on.
At the same time, OpenAI and Perplexity both announced AI shopping assistants that can scan products, compare options and tailor recommendations for you. Smaller retail‑focused AI startups told TechCrunch they are not freaking out yet, arguing that their deep focus on shopping, retailer integrations and specific niches can still beat general‑purpose giants trying to be everything to everyone.
When basic security still breaks
Away from the AI buzz, TechCrunch found a serious security bug in jury‑management websites used by courts in several U.S. states and parts of Canada. The software, made by Tyler Technologies, made it possible for attackers to guess juror login IDs and see very sensitive information like names, addresses, contact details and even personal answers from jury questionnaires before the company fixed the issue.
It’s a sharp contrast to the billions being poured into cutting‑edge AI infrastructure for governments. The story is a reminder that while everyone is racing to build the future of AI, some of the basic software that runs critical public services still fails on simple security hygiene, and ordinary people are the ones exposed when it does.
Startups wiring AI into the boring (important) stuff
TechCrunch counted 49 U.S. AI startups that each raised at least $100 million in 2025, which shows investor money is still flowing hard into AI. Most of that cash is going to teams that turn AI models into real products , things like infrastructure, developer tools and industry‑specific apps, instead of just launching another generic chatbot with no clear use case.
One example is Momentic, which raised $15 million to use AI to automate software testing for companies including Notion and Webflow. As AI‑assisted coding makes it faster to ship new features, tools like Momentic act as the quiet plumbing in the background, running huge numbers of tests so products don’t break every time developers push new code.
