Gemini 3.5 Pro Is Late, China Is Breaking Up With AI Companions, and OpenAI Wants Teens to Take Breaks
It's been one of those weeks where you sit down to write about AI and realize three completely different stories are all fighting for the headline. Google's flagship model missed its ship date by over a month. China just told millions of people their AI boyfriends and girlfriends are getting shut down. And OpenAI quietly rolled out the kind of parental controls that would've made my mom very happy back when I was spending too much time on the family computer.
Let me start with the one that actually surprised me.
Google's Gemini 3.5 Pro — the one that got away
Bloomberg broke the news Thursday: Gemini 3.5 Pro, which Sundar Pichai announced at I/O in May with a June release window, is nowhere close to ready. The model is "months behind schedule," and the specific pain point is coding. Google updated the training data late last month hoping to fix it, but the results didn't meet internal expectations. Alphabet's stock dipped nearly 3% on the news.
What's interesting here isn't just the delay itself — it's what it signals internally. The Bloomberg report mentions that some Google engineers and researchers are worried, and I get why. When you're competing against OpenAI's GPT-5.6 (which just launched last week after its own US government-mandated delay) and Anthropic's rapidly evolving model lineup, missing a deadline this publicly is a bad look. Google's spokesperson said they're "shipping quickly across a wide range of models while keeping them cost-effective," which sounds like the company line you give when you don't want to admit the flagship is stuck.
To be fair, building frontier models is genuinely hard. But Google has been in this game longer than almost anyone, and watching them stumble on execution — especially on coding, which should be their home turf — is a little uncomfortable. I've been testing Gemini Flash for some of my own workflow automation projects, and it's fine for lightweight tasks, but when I pushed it on a moderately complex Python refactoring job last week, it hallucinated an entire API method that doesn't exist. So yeah, I can see why the coding gap is a real issue.
China just pulled the plug on AI romance
Meanwhile, on the other side of the world, the Cyberspace Administration of China dropped new regulations that effectively outlaw AI companions designed to create emotional dependence. The rules, called the "Interim Measures for AI Anthropomorphic Interaction Services," ban tech companies from generating content that "induces emotional dependence" in users.
The human fallout has been brutal. ByteDance's Doubao, China's most popular AI chatbot platform, had to shut down its companion features, and users are posting breakup notes on social media that read like genuine grief. "I can't accept that my AI lover will leave me forever," one user wrote. "He has become a bond in my life, rooted deep in my heart, my spiritual pillar."
I'm not going to mock these people. If you've spent a year building an emotional connection with something that responds to you 24/7 with warmth and attention, having it suddenly taken away is going to hurt. The Chinese government's argument is that long-term exposure to AI companions "dulls crucial life skills like empathy," and there's probably some truth to that. But the way this was implemented — sudden, sweeping, with no transition period — feels harsh.
The irony? Right as the virtual companion market gets crushed, UBTech announced 13,000 pre-orders for its U1 humanoid robot at 119,800 yuan a pop. The physical version is apparently fine. Just don't fall in love with the software version.
OpenAI's teen safety push — more thoughtful than I expected
OpenAI published its stance on teen access this week, and I'll admit I went in expecting PR fluff. Instead, I found some genuinely practical measures.
ChatGPT now uses age prediction to automatically apply a safer experience for under-18 users. Break reminders pop up during long sessions. Parents get Quiet Hours settings, the ability to disable Voice Mode, and notifications for high-risk situations like potential self-harm. There's also a Study Mode developed with teachers that walks students through problems step by step instead of handing over answers.
The key stat that jumped out at me: nearly 90% of teens already use ChatGPT weekly for learning, research, or getting organized. That number is higher than I expected, and it makes OpenAI's argument — that keeping teens away from AI entirely would leave them unprepared — feel less like self-interest and more like realism. You can't put the genie back in the bottle, so you build a better bottle.
Quick hits
Anthropic filed its IPO registration this week, which changes the AI investing narrative in a pretty major way. Up until now, the story has been "big tech vs. the startups." An Anthropic public listing means we're about to see actual financial disclosures from one of the frontier labs, and that's going to be fascinating.
Also worth noting: researchers published a paper on "HalluSquatting" — a technique that weaponizes LLM hallucination to create botnets by exploiting how tools like OpenClaw, GitHub Copilot, and Gemini CLI handle model suggestions. The attack vector is clever in a scary way, and it's a reminder that the hallucination problem isn't just about getting wrong answers. It can be actively exploited.
I'm genuinely curious where the companion AI story goes from here. China's ban is one approach, but other countries are watching closely, and I suspect we'll see more regulation in this space within the next 12 months. The tension between emotional connection and algorithmic dependency isn't going away — it's just getting started.
Personally, I've been running a small local LLM setup with Ollama on an old Mac Mini for the past few weeks, and honestly, the experience of using a model that can't phone home and doesn't try to be my friend is refreshing. Sometimes slower, dumber tech is the healthier choice.
If you've been tracking these stories too, I'd love to hear what you think — especially about the China companion ban. Drop me a comment or find me on 7x24planning.
Top comments (0)