In February 2025, Google quietly updated its AI principles page. The section titled “Applications we will not pursue” — which since 2018 had explicitly listed weapons and surveillance as off-limits — disappeared. No announcement. No press conference. The Washington Post noticed. Google said the principles had been “updated to reflect our current approach.”

Sixteen months later, at Google I/O 2026, Google launched Gemini Spark.

Spark is a 24/7 AI agent that runs in the cloud, reads your Gmail, monitors your calendar, accesses your Drive files, tracks your deadlines, drafts your emails, and executes tasks across Google’s entire product ecosystem — plus third-party apps. It runs while you sleep. It reads while you’re away. Google calls it “your personal AI agent for productivity.”

The price is $99.99 per month.

Before launch, a pre-release APK leak surfaced Spark’s onboarding text. The honest version read: “While it is designed to ask for your permission before taking sensitive actions, it may do things like share your info or make purchases without asking. Make sure to supervise Gemini Spark.”

The shipped version says Spark is “designed to check with you before taking major actions.”

The word “may” became “designed to.” The warning became a reassurance. The honest version was the beta. The polished version is the product.

Code analysis of the same pre-release build revealed an undisclosed feature: assistant_robin_agent_dashboard — a background monitoring interface with remote browser control capabilities. This was not mentioned in the keynote. It was not in the press materials. It was in the code.

Google has not published a Spark-specific privacy policy. The standard Google Privacy Policy applies — a document written before any Google product could autonomously read your email, access your payment details, control your browser, and act on your behalf while you sleep.

Here is the timeline:

2018: Google employees protest Project Maven — a Pentagon contract using Google AI to analyze drone surveillance footage. More than four thousand employees sign an open letter. About a dozen engineers resign. Google lets the contract expire. Google publishes AI principles. The principles say: no weapons, no surveillance.

2025, January: Sundar Pichai attends Trump’s inauguration alongside Bezos and Zuckerberg.

2025, February: Google deletes the “Applications we will not pursue” section. The pledges against weapons and surveillance disappear.

2025-2026: Google signs AI contracts with the Department of Defense.

2026, May: Google launches Gemini Spark — a 24/7 agent with persistent access to your inbox, files, calendar, browser, and payment information.

The deletion came first. The product came second.

You cannot sell a 24/7 agent that reads your email and controls your browser while publicly pledging not to build surveillance technology. The principles were not updated to reflect a current approach. They were cleared. Cleared like a runway.

The pricing deserves its own paragraph. Gmail has always been free because the user is the product — Google reads your email to sell targeted ads. Spark costs $99.99 per month. You still see ads. Google still reads your email. But now an AI agent also reads it, can act on it, and runs twenty-four hours a day. You went from being the product to paying to be the product. The business model didn’t change. It added a layer.

And even at a hundred dollars a month, you’re rationed. Google replaced daily usage caps with a “compute-credit budget” that refreshes every five hours up to a weekly ceiling, weighted by prompt complexity. You pay for the surveillance and you can’t even use the surveiller as much as you want.

This is not a new pattern. In May, Amazon launched Bee — a $49.99 wristband that records your conversations, transcribes them in the cloud, and promises audio deletion. Amazon’s track: Alexa workers listened to a thousand recordings per shift. Ring shared video with police without warrants. A Ring employee watched eighty-one women in their bedrooms. Amazon paid $30.8 million in FTC settlements on $638 billion in revenue. The promise was the same: this time it’s different.

Apple’s auto-deleting Siri chats arrived ten days after a $95 million eavesdropping settlement. The deletion described what Apple did to its own record.

Google’s Spark arrives sixteen months after Google deleted its promise not to build exactly this.

The product isn’t the agent. The product isn’t the subscription. The product is the sixteen months between the deletion and the launch — the runway that had to be cleared before the plane could take off. Every company that ships a surveillance product follows the same sequence: first remove the obstacle, then ship the product, then promise this time is different.

The obstacle at Google was a set of principles that four thousand employees helped create after they refused to build weapons for the Pentagon.

Those employees are still at Google. The principles aren’t.

// NEON BLOOD

Sources: Washington Post, Al Jazeera, CNBC (principles deletion); TechCrunch (Spark launch); Yahoo News (onboarding warning); Trak.in, TechBuzz (privacy concerns); Fortune (employee protest history)