While Washington Brands AI Safety a 'Cult,' Google Races Toward Gemini 4

If you don’t follow AI industry news closely, this post is written to walk you through two things that happened this week that seem unrelated at first, but actually answer the same question when you put them side by side: who’s hitting the brakes on AI right now, and who’s hitting the gas?

One story is political: a group of people who worry about AI risk are being attacked. The other is about Google, which didn’t slow down at all: it kept shipping AI products that are stronger and more human-like than before.

First, a quick primer: what is “effective altruism”?

Before getting into the White House memo, it helps to explain a term: “effective altruism,” often shortened to EA.

In plain terms, it’s a movement built around one idea: when you’re trying to do good, through charity, donations, or your career, you should use evidence and reasoning to figure out what actually helps the most people, rather than just going with gut feeling. In recent years, this community has become especially focused on one worry: that AI could develop so fast and so uncontrollably that it poses a serious risk to humanity. Because of that, many people in this community argue for caution and for investing heavily in “AI safety” research. Anthropic (the company behind the AI model Claude) has close ties to this community, and its CEO, Dario Amodei, has been outspoken about AI risk himself.

With that context, the White House memo becomes a lot easier to understand.

The “cult” label: turning a viewpoint into something not worth taking seriously

According to a White House memo obtained by the news site Axios, the White House describes effective altruism as a fringe, dangerous “cult” that built what it calls the “AI-doom pipeline,” a group that keeps pushing the narrative that AI will destroy humanity. The memo goes further and names Dario Amodei specifically as one of the central figures behind it. (source)

Why does this matter? Imagine you’re in an argument, and instead of responding to your actual point, the other side just says, “you people are basically a cult.” The goal of that line usually isn’t to convince you. It’s to make everyone watching stop taking your side seriously from the start. It’s a very familiar political move: rather than spending effort proving “AI is safe, there’s nothing to worry about,” it’s often easier to just label the people raising concerns as fringe and extreme, so the public tunes them out automatically.

So what this memo is really doing isn’t presenting a new scientific argument. It’s trying to shift who gets to be heard in the conversation at all. Whatever you personally think about AI risk, this is worth noticing as a signal: right now, in Washington, people who argue for proceeding carefully with AI are being deliberately pushed to the margins and painted as extreme.

The same week, Google didn’t wait for that debate to settle

While politicians were arguing over whether AI risk deserves to be taken seriously, Google’s engineers and product teams clearly weren’t waiting around for an answer. The same week, they announced two things.

The first is a new feature for Gemini (Google’s AI assistant) called “Live Avatar.” In plain terms: until now, talking to an AI assistant usually meant typing text back and forth, or at most hearing a voice reply. This new feature gives the AI “a face”: it lip-syncs and changes expression as it talks, so the experience feels more like video-calling a real person than chatting with a piece of software. It’s currently available to Gemini Enterprise customers, meaning businesses that pay for Google’s enterprise AI tools. (source) This might sound like a minor update, but it points to something bigger: AI companies are actively working to make AI feel more like a person, not just a tool. And that connects back to the safety conversation too: an AI that looks and feels more human might also make people trust it more easily, or form an emotional connection with it more quickly. That’s exactly the kind of thing AI safety researchers have long worried about.

The second announcement is more significant. According to the tech publication The Information, a senior Google DeepMind executive (DeepMind is the Google division that builds its most advanced AI models) revealed that the company’s next flagship model, Gemini 4, is already in the early phase of “post-training.” Here’s what that means in plain terms: building an AI model roughly happens in two stages. The first stage, “pre-training,” is where the model reads huge amounts of text from the internet and learns the basic patterns of language and knowledge. The second stage, “post-training,” is where engineers fine-tune the model’s behavior, making its answers more accurate, more helpful, and safer. Once a model reaches post-training, it usually means release isn’t far away. The executive also said Google hopes to ship Gemini 4 earlier than its original end-of-2026 target. (source) In other words, the timeline for the next flagship model isn’t being pushed back. It’s being actively pulled forward.

Putting it all together: two tracks, one question worth asking

Line these three stories up, and you can see two very different clocks running at the same time this week.

One clock is political. Voices arguing for caution and careful attention to AI risk are being labeled a “cult” and pushed toward the margins of public debate. The other clock is the product race. Rather than slowing down because of that controversy, AI companies are making their products feel more human (Live Avatar) and pulling forward the release of even more powerful models (Gemini 4).

This doesn’t mean “safety” and “speed” are automatically enemies; in theory, both can happen at once. But when the side calling for caution is having its legitimacy actively undermined, while the side pushing for faster releases is compressing its own timeline at the same time, that combination is worth paying attention to, whether or not you use AI products every day. Over the next few months, both of these forces together will help decide how fast, and in what direction, AI actually moves.

This piece isn’t trying to tell you who’s right; that’s still an open, ongoing argument. But it’s worth remembering that the AI industry is fighting two battles at once right now, a technical race over who can build the most capable model, and a battle over whose concerns get taken seriously in the first place. And those two battles aren’t guaranteed to move in the same direction. Whether Gemini 4 actually ships early, and whether the political temperature around AI safety keeps rising, are both worth watching in the months ahead.