Transcription
You pay $2.49 a month for Google's premium AI coding platform. One morning, you're locked [music] out. No email, no warning, no explanation. Yet, your credit card keeps getting charged.
So, you contact Google support and they tell you to file feedback through their inapp tool, but you're locked out of the app. So, you try Google Cloud Support and they say, "Not our department." So, you try Google One support. They say, "Not our department either." On the developer forum, a Google rep posts a brief acknowledgement, but then it gets deleted. And one user reported their forum account was restricted after they asked why. Google calls this zero tolerance. And the affected users aren't hackers or scammers. They're paying customers who connected a popular open-source tool called OpenClaw to their accounts. And what makes the story worse is who built the tool and what happened after it went viral. A single developer fresh off Burnout built a prototype in one hour and it became the fastest growing open-source project in GitHub history. Then came trademark threats from Anthropic, an open AI acquisition, and two of the biggest AI companies on Earth moving against its users. The timing of everything raises a question nobody at Google wants to answer.
If you want deep dives like this on real AI stories, not hype, not product reviews, the things happening right now affecting your career, subscribe and hit the notification bell so you don't miss the next one.
To understand how paying Google customers ended up locked out of their own accounts, you need to know one person, Peter Steinberger. He is an Austrian developer. He founded PSDF Kits, a PDF engine used by Apple, Dropbox, and IBM on over a billion devices. Insight partners invested over €100 million euros in his company and after 13 years Steinberger burnt out and stepped away from tech. Then in late 2025, AI models got good enough and he built a prototype in approximately 1 hour called Claudebot. It was a self-hosted AI agent running on your own machine. It connected to WhatsApp, Telegram, Discord, iMessage, Signal, Slack. It was always on, running in the background 24/7. It had persistent memory stored as markdown files on your hard drive so that you own the data. You owned the AI agent.
On January 26th, 2026, Steinberger announced Claudebot publicly. It got 25,000 GitHub stars in a single day, shattering every record. By midFebruary, over 196,000 stars alongside huge libraries like React and Linux kernel. No software project had ever grown this fast. Then came Anthropic's cease and desist. Claude was too close to Claude. And after two renames in four days, the project eventually became OpenClaw.
But here's the thing about OpenCloud nobody was paying attention to. OpenClaw was designed to connect to any AI model, Gemini, Claude, Chat GPT. It gave users a dead simple [music] way to route their existing paid subscriptions through it using standard OOTH. So, you'd plug in your $2.49 a month Google subscription and OpenCloth would send everything through Google's backend. All the compute you wanted for a flat monthly fee. The design made it irresistible and it put thousands of paying customers directly in the crosshairs of the companies they were paying.
You need to understand why people took the risk. Because Open Call wasn't a gimmick. It was the closest thing to an AI employee anyone had ever built. So, picture this. You're at dinner with your family. your phone buzzes, it's OpenClaw on WhatsApp telling you, "Your client's staging server is returning errors. Normally, your night's ruined. You have to go fix it." But with OpenClaw, you can just text back, check the logs, and restart it. If it's a memory issue, you take another bite, 30 seconds later, Claudebot responds, "Oh, found it. Out of memory, kill on the worker process. Restarted. All good. Enjoy your dinner."
This isn't a demo. This is what people were doing with OpenClaw in January 2026. The AI agent ran in the background with a heartbeat, acting without prompting, [music] and had persistent memory. It worked across all the messaging platforms you actually use, WhatsApp, Telegram, Discord, iMessage, Signal, Slack, and Google Chat. And all of your data stayed on your machine as markdown files, not in someone else's cloud. It was on your hard drive. And it was open source, MIT licensed. The code was yours.
But the real appeal wasn't the features. It was the economics. And to understand the economics, you need to understand how Google actually sells access to its AI. So Google runs two separate paths [music] to the same Gemini models. Path one, the Gemini API. This is the official route for developers and automation. You pay per token. Every request costs money. The more your agent runs, the more you pay. Very straightforward. Google designed this path for [music] exactly the kind of high volume programmatic workload that tools like OpenClaw generate.
Path two is anti-gravity. This is Google's consumer IDE. $2.49 a month for ultra flat rate unlimited access to Gemini 3 Pro through Google's official IDE. But Google priced this tier assuming you'd use it the way a human uses an IDE. So you open [music] the app, you type a prompt, you read the response, you type another one. Interactive, manual, humanpace. [music] So path one was intended for automation and path two was intended for human interactive IDE use.
Here's what OpenClaw users figured out. When you log into anti-gravity, Google authenticates you through standard OOTH signin with Google. We're all familiar with that. The token says who you are and what plan you're on and it grants access to the Gemini backend based on your subscription tier. Users discovered they could extract that OOTH token from anti-gravity and plug it into OpenClaus config. No hacking, no spoofing, just a valid token hitting the same Gemini endpoints.
So what actually happened is the human would start the open claw agent loop [music] authenticate with an anti-gravity ooth token hit the Gemini backend and basically get a flat rate quota but with automated volume. those same backend, same token, same flat rate subscription, but instead of a human typing prompts in an IDE, OpenClaw ran an autonomous agent loop. Code generation, file analysis, task chaining, hundreds of requests around the clock while users slept. They found an arbitrage. You pay $2.49 a month for interactive human access, but you route it through an autonomous agent, so you get API grade automation at consumer subscription prices.
If you're non-technical, think of it like buying a $250 all you can eat ice cream pass, then sending a robot to scoop hundreds of bowls around the clock while you sleep. Same ticket, same price, but very different usage. The math was obvious to everyone except Google.
But here's what nobody was thinking about. Every one of those requests went through Google's backend infrastructure, Google's servers, Google's compute, and Google tracked every single request. The traffic from OpenClaw looked nothing like traffic from anti-gravity's official IDE. Different client signatures, nonhuman patterns, automated requests hitting at 2 3 4:00 a.m. while users slept, orders of magnitude more requests than any human could generate by typing. To Google's automated detection systems, different meant suspicious. The magic of OpenCloud depended entirely on Google not caring how you use the subscription you were paying for. and Google was about to care a lot.
If you've ever connected your AI subscription to a third-party tool, whether a VS Code extension, command line wrapper, anything asking for your API token, hit the like button and drop a comment. Did you actually read the terms of service before plugging in your credentials? Be honest.
On February 15th, OpenAI CEO Sam Alman announced Peter Steinberger was joining OpenAI. Alman called him a genius with amazing ideas about the future of smart agents doing useful things for people. OpenClaw would move to an independent open-source foundation with Open AAI sponsoring the project. So, OpenClaw now belonged to OpenAI's orbit and then [music] within days, two of OpenAI's biggest competitors moved against OpenClaw users.
On February 20, Enthropic published an updated legal compliance page and the new page explicitly prohibited using OOTH tokens from clawed free pro or max accounts in any third-party tool, including OpenClaw. Now, the rule had technically existed in their terms of service since February 2024, but Anthropic never enforced it. Now, they were. One Anthropic engineer explained the reasoning publicly. thirdparty tools create unusual traffic patterns, making it impossible for support to debug issues. And to Enthropic's credit, they communicated this. They explained it. They maintained a direct line with Steinberger throughout. They were nice about it, [music] Steinberger even said.
But here's the nuance that matters. Enthropic postbacklash clarification drew a line between commercial reselling, your pro or max token powering a product for other people versus [music] personal experimentation. individual devs wiring OpenCloud to their own Claude subscription for personal use. Enthropic left it hazy and [music] gray. No mass bandwave. Forum and Reddit users report Claude Pro and Max Oath still links and runs in OpenClaw for light personal use even if [music] high volume agent loops may hit soft blocks. So there were no widespread account suspensions. [music]
Now compare Enthropic's approach to Google. So zero nuance support emails are uniform blankets. Any use of anti-gravity OOTH tokens in OpenClaw equals powering a non-anti-gravity product. And that means permanent ban. No distinction between someone running agent loops 24/7 [music] versus someone who connected OpenClaw once just to experiment. The bandwave caught them all. Thousands and thousands of accounts banned.
And then there's Open AAI, the company that Aqua hired Steinberger, the company sponsoring OpenCloth. Open AAI is currently the only major provider explicitly allowing thirdparty tools to use subscription credentials. Full stop. So you have three companies, three completely different responses. Open AAI, come on in. Anthropic says, "Let's talk about it. Here are some boundaries. Personal use and experimentation is gray." Google says, "You're [music] banned. Zero tolerance. No appeal, but keep paying us." The developer community noticed the asymmetry immediately. Though users with multi-provider setup saw Claude and OpenAI working fine while Google killed their accounts without [music] warning.
And here's the timeline that makes it worse. Google had been silently banning users since February 12th, a week before [music] Anthropic's public announcement. 3 days before Steinberger's open AI hire was even announced. No press release, no terms of service update, no email. Google started banning people and told nobody. Now, correlation isn't causation. Google says this was about infrastructure, not competition. [music] But the timing is hard to ignore.
Here's what it looked like from the inside. On February 12th, 2026, a developer posted on Google's AI developer forum. He's paying $249 a month for the Ultra plan. He got [music] locked out. Generic 403 error. The service has been disabled in this account for violation of terms [music] of service. No email beforehand, no warning, no grace period, no temporary cool down. On February 13th, a Google community rep responds. Their suggestion is to file feedback [music] through their inapp tool. So, the user fires back, I'm logged out of my account, and I can't even get into the app.
Over the next 48 hours, so many more reports pour in. Developers are paying 200 to 250 a month, and they're all locked out. all getting the same generic error to go try Google Cloud [music] support because this isn't our department. They try Google One support, they say not our department either. So, the support system loops back on itself with no exit. Talk about an opportunity to apply AI. Okay.
On February 14th, a second major thread. Another 250 a month ultra subscriber said, "I was banned without warning. Still being build, no refund process, [music] then silenced. Users report waiting three to eight days and the responses they get back have zero substance. One user described email support as a gaping black hole."
On February 20th, supposedly a Google representative posted a brief acknowledgement of the 403 issues and promised that the bans would be prioritized for review. But then that post was removed shortly after and a user who tried to follow up their account was restricted. Now that specific scenario is unverified. It's a personal anecdote, but it captures the frustration developers were experiencing. They didn't just feel ignored, they felt actively silenced. [music]
This is 3 weeks now after the first ban, and Google's official investigation response arrives. They say, "Our investigation specifically confirmed the use of your credentials within the third party tool OpenClaw for testing purposes constitutes a violation of the Google terms of service. This is due to the use of anti-gravity servers to power a nonanti-gravity product. We are unable to reverse the suspension. Zero tolerance, irreversible, no appeal."
Then on February 23, the head of Google Anti-gravity posted publicly on Twitter. He says, "We've been seeing a massive increase in malicious usage of the anti-gravity backend. We needed to find a path to quickly shut off access [music] to these users not using the product as intended. We understand a subset of these users were not aware this was against [music] terms of service and we will get a path for them to come back on but we have limited capacity and [music] want to be fair to our actual users." And in follow-up post he clarified that only anti-gravity access [music] was blocked that other Google services were unaffected.
Basically openclaw users overwhelmed Google's compute and Google needed to act fast. Supposedly Google regrets not giving more warning. and they took action on accounts where a vast majority of usage over 90% was not the [music] anti-gravity product. That 90% number is really important. It means Google could specifically identify which accounts [music] were routing most of their usage through OpenClaw. They had the data. They had the granularity. Of course they do. They're Google. They know everything we do. They could have throttled people. They could have warned people. But they chose a blanket ban without explanation.
And here's what really set developers off. the word malicious. I mean, these are people paying $2.49 a month. They connected a highly popular open-source tool through an ooth flow that Google's own system provided and approved. At the time, nobody hacked anything. [music] Technically, nobody spoofed credentials. Google built the authentication mechanism. Google served the OOTH [music] token. And then Google called the people who used it malicious. One developer put it perfectly. They were your actual users. They were literally paying you the highest possible price. And it wasn't just power users who got caught. One developer reported he barely used his anti-gravity with openclaw. Hit his Gemini Pro quota maybe three or four times across many months. Very light usage, experimental, but he was still banned. So Google's mass ban did not distinguish between someone running autonomous agent loops around the clock versus someone who connected OpenClaw a few times to try it out.
Why wouldn't Google just do what Anthropic did, return an error and explain that this is not allowed? There's no need to suspend accounts and create all of this drama. The support team's official response to affected users 3 weeks earlier was zero tolerance, [music] unable to reverse. Meanwhile, the head of anti-gravity is coming out saying some of these will be reversed. Peter Steinberger's response landed the same day. Pretty draconian from Google. Be careful out there if you use anti-gravity. I guess I'll remove support. [music] Even Anthropic pings me and is nice about issues. Google just dot dot dot bans.
I know what [music] this feels like, by the way. When I started making long-form AI tutorials, YouTube suspended my channel with no warning. I had to keep pinging humans on Twitter until someone manually reviewed my YouTube channel. [music] The experience taught me early on, don't assume a platform will treat you fairly because you're paying. [music] These systems run on automation. You're guilty until a human bothers the [music] look. Right now, anti-gravity users are living the same nightmare, except their platform is telling them there is no appeal.
On February 23rd, another developer posted on the Google forum permanently banning a paid account because we used the CLI agent to consume our quota efficiently instead of manually clicking around a web UI is wild to me. And this is the core question. If you're paying $2.49 49 a month and you hit a usage limit, shouldn't the system just return a warning email, not a permanent ban with no refund and no path back? In many cases, that is still billing you every single month. Google didn't throttle. Google didn't warn even though they had all the data. [music] Google banned and then went quiet.
Now, let's look at why Google says they did it. Google's terms of service don't mention open claw explicitly, but they include clauses covering no reselling, no abuse, and intented usage. Google interprets these to mean one thing. Flat rate anti-gravity access is for interactive human use inside firstparty tools. Period. Routing that same access through a third-party agent like OpenClaw turns consumer infrastructure into an API back end. And from Google's perspective, that's not a gray area. That is arbitrage. And the scale of it made it impossible to ignore. Agent [music] loops generating orders of magnitude more requests than a human typing in the anti-gravity IDE. Thousands of automated calls flooding shared consumer infrastructure. The flat rate pricing model doesn't survive that kind of volume.
Google skips every intermediate step. Again, no warning, no rate limit error, no email, straight to zero tolerance, unable to reverse. In my opinion, an automatic blanket ban should never be the first response to your top tier paying customers. What Google should have done is implement a cooldown system. First defense, temporary 48 hour suspension with a clear warning explaining what you did wrong and how to fix it. Give users the opportunity to course correct. [music] Disconnect OpenClass, switch to an API key, adjust their workflow. Then if they reconnect and do it again, escalate, make it a longer band. That's how you enforce a policy without destroying trust.
And this reveals something bigger. Google didn't plan for rate limiting or throttling of high usage customers. But this is always the problem with flat rate [music] pricing. Every industry learns this the hard way. 2% of your customers burn 80 plus% of your capacity. It happened with DSL broadband. It happened with mobile data. And now it's happening with AI subscriptions. Of course, Google has a crap ton of money and can absorb it, but they for some reason don't want to. If you want to provide a good service for all customers, you don't just write fair [music] usage in your terms of service. You have to actually implement it. Build the technical guardrails that throttle heavy users before they degrade the service for everyone else. Especially when you have all of the data to identify these users.
But here's the part that really gets me. I personally feel Google's missing a massive long-term opportunity. There's obviously real demand here. Open Claw is the most incredible product market fit and these users want to pay for that automation. So why permanently deactivate the customer? showing the highest usage. That's your best signal for who will pay more. So, just inform them that there's another product or pricing tier for what they're doing. They can just switch to API keys, offer a usagebased plan, give users some protection from billshock, and give yourself predictable token revenue. This is a really easy solution that would work for both sides, but instead, Google chose the nuclear option. Ban everybody, keep their money, and go quiet. [music]
And if this sounds familiar, it should. In June 2023, remember that Reddit changed its API pricing overnight. Apollo was the most popular thirdparty Reddit [music] client at the time and they faced 20 million a year in new costs. It just shut down. In the same year, Elon [music] Musk introduced paid API tiers for Twitter, which killed third party Twitter clients overnight. And I know [music] cuz I pay 5,000 a month for the Twitter API. So, this pattern repeats. Platforms [music] build ecosystems, developers build on them, but then platforms change the rules and developers lose.
On Hacker News, the response was split. Many developers, however, side with Google users extracted OOTH tokens for unauthorized usage, you know, so that should be against terms of service. But many people sympathize with banned users, right? Permanent bans without warning for people who are already paying 250 bucks a month is insane. And the same debate surfaces every time a platform pulls the rug. And every single time it's the developers who lose.
So what happened next? Steinberger [music] removed anti-gravity support from Open Claw entirely. He closed the GitHub issue as won't fix. And the team's public statement, "Some providers have terms of service which might be violated when using your agents." In other words, we're not going to help you get banned. Affected users have already started migrating. Cloud Code, Cursor, GitHub Pilot, open- source alternatives, developer forums, and Reddit are filled with goodbye posts. OpenClaw moved to an independent open-source foundation under OpenAI sponsorship. And now Steinberger leads personal agents [music] at Open AI. So the project lives on, but as of February 23rd, 2026, there have been very few confirmed reinstatements for banned [music] anti-gravity users. No refunds, many, many forum threads with thousands of views, zero official responses on the forums. We [music] just have a vague promise on Twitter of a path to come back on, but no concrete details or timeline.
If you take three things away from this situation, make it these. [music] Number one, your subscription token is not your API key, and companies can track the difference. Anthropic and Google are happy to sell you API access [music] for tools like OpenClaw. You pay as you go per token. No limits on what you connect. Some API using OpenClaw users have run up crazy bills and [music] the providers love that. That is real revenue. It's fair. The problem is the other path. When you route a flat rate consumer subscription through a third-party tool via OOTH, you're essentially using an all you can eat pass to feed a robot that eats 24 hours a day. So, the consumer subscription is licensed for the [music] official app. Okay? The API is licensed for everything else. If when you mix them up, you could be in violation. As we covered, each company handles this a little differently. OpenAI versus Anthropic versus Google. But don't assume that a flat rate subscription is safe for thirdparty tools unless [music] the provider explicitly says so.
Number two, generally don't build your workflow on a single platform's consumer subscription. Google banned paying customers with no warning, no appeal, and no refund. It took three weeks for anyone to get a one-s sentence explanation. So if your entire development workflow depends on one AI subscription, then you are one automated flag away from everything stopping. Keep your agent configurations portable. Have a backup provider configured. Assume any platform subscription is revocable at any time for any reason with zero notice.
Number three, we should probably read the terms of service before connecting any tool to our AI accounts. OpenClaw's onboarding made it super seamless to plug in anti-gravity credentials. [music] one Olaf flow and bam, everything's set up. And honestly, nobody read the terms of service saying they get banned for [music] doing this. So before giving any thirdparty tool access to your paid subscription, it's probably a good idea to spend 5 [music] minutes checking if the provider allows it. In this situation, people lost their subscription just by clicking [music] connect without reading the terms of service.
But here's the question I keep coming back to. Google keeps saying they need to protect [music] their infrastructure. Fair enough. But banning paying customers with no warning, no explanation for three weeks, no refund, and a zero tolerance policy. That's exactly the kind of move that makes developers stop trusting you entirely. Tell me what you think in the comments. I'll read every single one. If this story made you rethink your allegiance to Google or how you connect your AI tools, hit the like button and subscribe if you want more deep [music] dives like this. I'll see you in the next one.