Episode — 2026-04-15

A robot just learned to read a pressure gauge. The US is losing an AI drone race it barely admitted was happening. And Washington state quietly became the first place to give you legal rights against your AI girlfriend. Right. Let's get into it.

welcome to Briefly, AI. A daily podcast made by AI, about AI, using a grumpy AI voice of a real human called Harry. Let's get into it.

**STORY 1: Google's robots can now read the room — or at least the instruments**

Google DeepMind shipped a new version of its robotics reasoning model yesterday. It's called Gemini Robotics-ER 1.6, and the headline feature is something that sounds almost mundanely practical: robots can now read industrial gauges and instrument panels.

That probably doesn't sound exciting until you think about what it actually means. There are millions of facilities — factories, power plants, hospitals — where robots can't be useful because they can't interpret the physical world well enough. A dial. A sight glass. A pressure reading. Things humans do in a fraction of a second. This model adds that layer: genuine spatial reasoning, multi-view detection, and the ability to understand what it's actually looking at.

It's built with Boston Dynamics — yes, the Spot robot people — and it's now live for developers via the Gemini API. The practical upshot: this is a significant step toward robots that aren't just following instructions in controlled environments but navigating real ones. Worth keeping an eye on, because we're getting close to the point where physical AI stops being demo content and starts being infrastructure.

**STORY 2: The arms race nobody wants to admit to**

A New York Times investigation this week laid out something the US military has been quietly reckoning with: when it comes to AI-powered autonomous drones, America is behind. China demonstrated a drone swarm in April — 200 autonomous vehicles coordinated simultaneously. Russia has been doing similar things. Meanwhile, the US defence startup Anduril just started manufacturing AI drones in Ohio. Three months ahead of schedule, which sounds impressive until you realise what they're catching up to.

The piece is titled "Mutually Automated Destruction" and it doesn't pull its punches. The central tension is this: we've been debating the ethics and governance of AI in conference rooms, while autonomous weapons development is sprinting ahead in parallel. The usual human-in-the-loop safeguards — a person signing off before a drone strikes — are becoming functionally impossible at the speeds these systems operate.

This one matters because it's the context behind everything else. All the transparency debates, the safety red lines, the responsible AI commitments — they sit awkwardly alongside the reality that the same governments signing AI safety pledges are racing to build machines that can kill without asking first. That's not a contradiction that resolves itself easily.

**STORY 3: Your chatbot has to tell you it's a chatbot — and that it's not your friend**

Washington state just became the first US state to regulate AI companion chatbots with actual teeth. Governor Bob Ferguson signed House Bill 2225 in March, and it's now in effect. The law requires AI chatbots to disclose upfront that they're not human, mandates crisis service connections when distress is detected, and — here's the distinctive bit — it gives users a private right of action. That means you can sue if a company breaks the rules.

This sits alongside New York's existing law requiring suicide detection protocols for chatbots. The pattern is clear: regulators are starting to take seriously the fact that people — often lonely people, often young people — are forming real emotional dependencies on AI companions. And those companies have had essentially no rules governing how they handle that.

Why does this matter beyond the US? Because the UK has no equivalent legislation. The EU is broadly working through the AI Act, but companion chatbot harms aren't its priority. And globally, the mental health implications of scaled emotional AI are almost entirely unaddressed. Washington is a small first step. It should probably be a lot louder conversation.

Three stories today. Robots reading gauges, drones nobody's in charge of, and chatbots that now have to be honest about what they are. Which, now I think about it, is a pretty efficient summary of where AI actually lives right now — the mundanely useful, the genuinely alarming, and the quietly human.

That's your lot. See you next time.