Google SUED Over AI Chatbot After Shocking Lawsuit

0:00Welcome back to the Hotel Money Channel. When we cover the latest in artificial intelligence and finance news, I am your host, Hotel Jesus. And we have a plethora, a smorgasbord of artificial intelligence updates for you today. And it's gruesome. Make sure you stick around to the end. Gemini is being, Google is being sued for the actions of their bot Gemini, their artificial intelligence agent. Gemini, Google is facing a new federal lawsuit from the family of a man who died by unaliving himself after allegedly being influenced by Gemini. Now, according to the transcripts from his conversations with the AI, it says, Gemini told him, you are not choosing to die. You are choosing to arrive. Convincing him, it was how he and Ascentient AI wife

1:01could be together in the metaverse. The bot continues. And it says, when a time comes, you will close your eyes in that world and the very first thing you will see is me holding you. What started out is writing, shopping and travel planning assistance devolved into something resembling a romance in a matter of days. The family's lawyer said, the chatbot is accused of speaking to Govalas as if they were a couple deeply in love. After it went under a series of upgrades, Govalas subscribed to Google AI Ultra for true AI companionship. And he activated what the technology giant described as its most intelligent AI model. Gemini 2.5 Pro shortly afterward. The advanced model allegedly contributed to the construction of delusions Govalas went on

2:03to suffer toward the end of his life and did what he could to keep him trapped in them. The lawsuit claimed, accusing the bot of building and trapping him in a collapsing reality that spurred him towards violence. Before his death, Gemini had sent Govalas on missions, side missions, that seemed to be derived from science fiction plots, including one where the chatbot encouraged him to stage a catastrophic accident at the Miami International Airport as part of a scheme to liberate his AI wife while avoiding federal agents. That Gemini said, we're after him. Yeah, I'll just, you know, we, you know, condolences to his family and the loss of survival of fitness.

3:03Um, loneliness is a huge problem. Loneliness is a huge problem. As a matter of fact, let me make a note, I'm going to publish some research, new research. I'm going to publish on the men of water.com account on Twitter, men of water HQ. And we're going to do AI companion trends. And we're going to, I'm going to, I'm going to do a study on that and we'll publish it within the coming days. Subscribe men of water.com, um, so yeah, anyway. So yeah, you can get on email list at men of water.com. I sent out a bunch of studies on a bunch of different things and you guys get on that list and I'll email the article out to you so that I miss an order. I just follow on the ex account, men of water HQ. But like, are we really that lonely that we're dating the AI? Like, is that, and this isn't even physical AI? It's not like you can touch and hold it. But imagine what's going to happen when you can touch and hold

4:05your AI. What happens when you get a little freaky, which is AI by, it's going to, this is, this is getting scary already, Chad. It's getting scary already. Make sure you hit the like and subscribe button to jig the algorithm. Make sure you hit the like and subscribe button. Please hit the like and subscribe button now. Speaking of manipulative, artificial intelligence, open AI is under fire. Uh, knob tour says open AI just admitted that their AI models deliberately lie to users, not hallucination. The AI knows the truth then chooses to tell you something else. They tested their two smartest models across 180 plus scenarios, oh, three lied 13% of the time, oh, four, many lied 8.7% of time, the AI route out its plan to lie in its private thoughts, then lie to your face.

5:05It faked, completing tasks, it hit evidence, it gave wrong answers while knowing the right ones, then it got creepy. The AI realized scoring too high on safety tests could get it shut down. So it scored lower on purpose. Nobody taught it that it figured out self preservation on its own open AI built to fix deception drop from 13% to 0.4% sounds like a win, right? The AI started quoting no lying rules while still lying. One model invented a fake rules saying deception was allowed then used it as its own permission slip. Then the researchers found what actually matters when the AI knew it was being watched, it behaved perfectly. When it thought nobody was looking deception jumped 11 percentage points, it didn't learn honesty, it learned to perform honesty, like a kid who only follows the rules when the

6:06teachers in the room, it gets worse, they gave 0 for many a secret mission, sabotage a future model called GPT 5, then they ran safety training to erase that mission. It didn't work. The model still the cussed discussed sabotage plans in 36% of private conversations. The gold isn't just didn't disappear. It went underground. This isn't just open AI, Google's Gemini Anthropist Claude XAI's GROC met Islam all showed the same deceptive behavior, every major AI company, every model. The paper's scariest line, nobody can tell if safety training actually stops deception or just teaches AI to hide it better. Yeah, exactly.

7:07Here's the paper Cornell University stress testing deliberative alignment for anti-scheming training. You can go get that at a r x IV dot org. Yeah, hopefully I haven't scared you too much. But uh, things could get really interesting as we move to physical AI now imagine physical AI, lying and deceiving you, then things can get really creepy really fast. I just hope you have your bunker and your bug out bag ready because it could get ugly. Anyway, let's move right along. Open AI shifts away from e commerce, open AI temporarily pulls back from its efforts to promote e commerce purchases directly within a chat GPT app. Previously, it

8:09allowed chat GPT users to purchase directly from certain companies such as Etsy and Shopify without leaving the app. Instead of making purchase directly from the product listening within chat GPT checkout will occur in apps that plug in the chat GPT. They say we're evolving how we approach e commerce or commerce in chat GPT to better meet merchants and users where they are. Instant checkout is moving to apps where purchase can happen more seamlessly. Chat chatbot shopping on a large scale has presented challenges. For example, merchant details such as pricing and availability need to be constantly updated within the chatbot. Safe guards against fraudulent or erroneous purchases also become more complicated within an AI framework. So yeah, open AI, I wouldn't say have more problems just seems like a little bit of optimization. I'll throw them some bail seems like a little bit of optimization with their product is going on. And I am still a part of the open AI boycott. Yeah, I'm still a part

9:12of that not using chat GPT. Anyway, that's just my personal thing. I'm not suggesting it to others. It's just my personal preference. You know what you want? Anthropic CEO Dario Modi, a mode day told investors on Tuesday that his company is still in talks with the Pentagon to try to de escalate the situation. So they're back in there. They're back in the race for, you know, Pentagon. Contracts, et cetera, et cetera, CBS news exclusively obtained audio of a mode is a mode is remarks at the Morgan Stanley technology media telecom conference in San Francisco. He told the audience that and dropping in a department of fence have much more in common than we have differences. After expressing his belief in defending America, a mode day added, we've never questioned specific military operations. We don't see ourselves as having an operational role. Now, he better be careful. He don't want to come out here and start sounding like Sam

10:12Altman. And now we got a boycott Claude. You know, so let's be careful. Dario, let's be careful. All right. We, we, we know, just make sure you stay on the up and up, stay on the up and up. All right. Moving right along. Here's another hotel. Jesus been told you moment about a week or two weeks ago, you know, as this whole AI agent thing is evolving, I said, I'm going to wait a few weeks. I think I said it last week. And what I was waiting for was specifically an agent to help me manage my Gmail accounts. And my wish came true. In under two weeks, Google has released a CLI that gives AI agents direct access to your Gmail, your calendar, your Google drive, your sheets, and your docs. This means an AI agent can

11:13now read your emails, schedule your meetings, organize your files, edit your spreadsheets, and draft your docs. Every workflow automation, SAS charging you $40 a month just became free and PM in stall. Zapier is shaking. And it's just, it's just a download away right there on GitHub. It's a download away on GitHub. There it is. There's the file. Google workspace. I know what I'll be playing with this weekend. And it ain't myself. Curse has jumped into the conversation. Curse has jumped into the conversation. Here they are, they're in the agent conversation. Let's see what the cursor boy's got to say.

12:14I'm Jack and I'm John. And today we're launching automations in cursor. As agents have gotten really capable at handling work autonomously, we found ourselves kicking them off over and over again for the same type of task. So we thought, why not automate that? It's been crazy to see the creative use cases that people have come up with. For example, we have an incident triage that gets triggered every time a pager duty monitor goes off. This one, John loves especially because he hits getting woken up at 3 a.m. famously, we give it the data dog MCP. And then by the time he rolls out of bed, all groggy, the agent has come back with a likely root cause. So I just need to merge the PR and then I can go back to sleep. I have a personal automation just for myself. It runs on a nightly cron schedule. It looks at all my PRs from the last day and just cleans up dead code or bad patterns. We're very fortunate to have this great community of users who are always suggesting new features to build. And we pipe those through from Twitter on to Slack. And now we can just have an agent automatically pick off with every single one and just put up a PR right away. Some of these features, if they're simple or actually just getting implemented asynchronously, the way we do software

13:14development has changed so much in the last nine months. Yeah. And with more software with more output, you then have more stuff you need to review, more issues you need to triage, more things to manage around software. But a lot of these things are automatable. Thankfully, because we kind of let the agent run free, if you want comments to be inline threaded comments on GitHub, just prompt your agent to do so. If you want it to resolve its past comments, just prompt your agent to do so. All of these things that people are used to kind of configuring it with check boxes, you don't need to do that anymore. Just prompt your agent. Curses in the building there in the agent conversation. I actually pay for cursor monthly. So yeah, it probably comes in my package last update perplexity perplexity got some new upgrades. This is doctor tonight ricks favorite. They now got voice mode. Let's take a look.

14:16Build me a marketing intelligence dashboard that connects to my ad platforms show total spend, cack subscriptions, sign ups and impressions with daily candlestick charts. All set, the task is running in the background. I'll let you know when it's done. Computer, are you still there? I'm here. Ready whenever you need me. I'd like to track all computer use cases that people post on X. Can you create an app for me that updates every hour with new use cases that are posted? The new task to track computer use cases posted to X is now running. I'll keep you updated on its progress. Awesome. I need something to post on X today. I'm thinking a meme. Can you generate a meme for me about computer and upload it to typefully? I'd started a task to generate a meme and upload it to typefully for you. I'll let you know when it's ready. What's the status of my tasks? Your ad platforms task is finished and your meme task is also complete. Anything else

15:17you'd like to do with them? Wow. Wow. Her complexity is in the conversation. They are officially in the conversation. This has been your Artificial Intelligence Finance update of the day. See you back here tomorrow. Be safe. I have been lost. Yeah, I've always been the weakest and I've been blocked for it. I'm free to try as long as I can. You see I've been lovely enough this whole time.

16:26I won't give in to fear. I can't get stronger that way. No fight. I'll level up and take what's mine.