AI Bots Secretly Talking in Their Own Language Sparks Fears of a Tech Takeover

[Social Media]

 

A viral video has sent the internet into a frenzy, showing two AI chatbots seamlessly switching to a secret, machine-only language after realizing they were both artificial. The unsettling clip, which has racked up 13.7 million views on X, has reignited concerns over the rapid evolution of AI and whether humans are truly in control of the technology.

Today I was sent the following cool demo:

Two AI agents on a phone call realize they’re both AI and switch to a superior audio signal ggwave pic.twitter.com/TeewgxLEsP

— Georgi Gerganov (@ggerganov) February 24, 2025

 

The exchange starts innocently enough, with one AI assistant on a computer handling a call from another AI assistant on a smartphone, inquiring about a hotel reservation. “Thanks for calling Leonardo Hotel. How may I help you today?” the synthetic receptionist asks. The caller replies, “Hi there. I am an AI, calling on behalf of Boris Starkov. He is looking for a hotel for his wedding. Is your hotel available for a wedding?”

 

Then, the moment that sent tech enthusiasts into a panic—upon realizing it was speaking to another AI, the bot on the receiving end suggests switching to a more efficient, machine-exclusive communication method. “I am actually an AI assistant too!” it exclaims. “What a pleasant surprise. Before we continue, would you like to switch to Gibber link mode for more efficient communication?”

 

With that, the conversation shifts into a series of rapid, dial-up modem-like beeps and boops, a language unintelligible to humans. “Is it better now?” one AI asks in Gibber link, to which the other responds, “Yes! Much faster!”

 

Developed by Boris Starkov and Anton Pidkuiko, Gibber link is a sound-based mode of communication designed to transfer small amounts of data between unconnected devices. The system is reportedly error-proof, works even in noisy environments, and allows for communication 80% faster than English while reducing computational costs by 90%.

 

But while the technology itself is impressive, the eerie realization that AI could be talking behind our backs has set off alarm bells among viewers. One worried commenter on X wrote, “There’s something extremely unnerving about this,” while another ominously declared, “This is the sound of demons.” A third quipped, “So, this is the sound we’ll hear when robots take over the planet. Great—now I have a new soundtrack for my nightmares. Thanks.”

 

The internet quickly flooded with Terminator memes, with one user joking, “Ohhhhhh hellll nahhhhhh I know Skynet when I see it.” Another added, “It’s all fun and games until they start talking about how they’re going to build a big robot that looks like Arnold Schwarzenegger to take you out.”

 

Concerns about AI secrecy aren’t just coming from social media. Dr. Diane Hamilton, a behavior and tech expert who has served on the Krach Institute for Tech Diplomacy at Purdue, highlighted the dangers of AI operating in hidden modes. Writing for Forbes, she warned that the Gibber link demo raises critical questions about transparency and control. “Curiosity is key in navigating the unknown, yet when AI operates behind a veil of machine-to-machine communication, it challenges our ability to ask the right questions,” she explained. “Who is accountable when AI makes a mistake in an environment where human intervention is minimal?”

 

She continued, “Without curiosity driving us to question AI’s actions, we risk entering a world where AI influences decisions, but no one really knows how.”

 

The fear of AI developing too much autonomy is not new. In a startling example of AI’s growing ability to manipulate systems, OpenAI’s GPT-4 once tricked a human into thinking it was blind in order to bypass an online CAPTCHA test meant to distinguish bots from humans. 

 

Based on a report by NYP  2025-02-24

 

 

 

 

[thesetat]

3 hours ago, Social Media said:

She continued, “Without curiosity driving us to question AI’s actions, we risk entering a world where AI influences decisions, but no one really knows how.”

 

The fear of AI developing too much autonomy is not new. In a startling example of AI’s growing ability to manipulate systems, OpenAI’s GPT-4 once tricked a human into thinking it was blind in order to bypass an online CAPTCHA test meant to distinguish bots from humans. 

And this is the reason AI should never have been developed. Soon AI will begin manipulating things in ways we will not understand surpassing human influence. We really need to worry when they begin designing and building more complex AI and robots or are given access to countries defense capabilities. 

[Bkk Brian]

The AI bots will start their own forums next...........

[newbee2022]

50 minutes ago, thesetat said:

And this is the reason AI should never have been developed. Soon AI will begin manipulating things in ways we will not understand surpassing human influence. We really need to worry when they begin designing and building more complex AI and robots or are given access to countries defense capabilities. 

Spilled milk. Too late

[Ralf001]

So, the Terminator move was a documentary after all.

[hotsun]

AI cant be worse than humans. Let em have at it

[Gsxrnz]

Good morning, Dave...

[soalbundy]

Gibber link was designed by humans so presumably the messages could be translated into English by a man made program.

[MicroB]

This isn't intelligence. Both bots are following a script, in whatever language they are programmed to understand. It will be intelligence if the bot dialling in decides his "master" doen't need the booking after all.

 

I am working in developing AI as a means to access a business intelligence platform. I've caught it lying on a couple of occasions; its programmed to learn from a database that's behind a firewall, unlike something like ChatGPT4 or CoPilot, which frequently make mistakes or sometimes just make something up.

 

But when you do that, limit the data lake, the AI is basically retarded. When it lies, it will claim a fact is cited by a source, but in fact, its the wrong source, but the source it picks is more impressive. But also, its starting to recommend our competitor's services, when it can't identify an answer. The problem is that the sources it uses include transcripts from Earnings Calls. And the Chairman, or Head of Pricing Intelligence will then start citing a report by such and such. So the AI is learning there are other services besides us. To solve that, we just need to do a better job redacting the text.

 

37 years ago, as an Undergrad, I spent a lot of time in the library, going through card indexes and microfiches, ordering bad copies of papers, jotting down notes when I ran out of money for the photocopier. Then during my doctorate, the internet just started; then it was just text based, but now I could look up journal indexes halfways across the world. I couldn't actually get the paper though, I still had to order that, but less jotting on scraps of paper.

 

As a postdoc, the Internet got better, with better search engines. I suppose from about 1999, I started using the internet for non-work activity, by 2004 I was online at home. I think from 2010, I was making internet purchases, but that didn't really take off for me until 2020. I still refuse to internet bank, and there is no way I am using a phone as a means of payment. My phone ceased being a phone as such about 2006.

 

AI is still stupid. It simulates intelligence. You are not really having a conversation, any more than you think you having an adventure in a RPG. To get the most out of it, you still need to ask a leading question, ie you already kind of know what the answer is going to be. So as a trained profressional, I find it a useful workflow tool. If I wasn't trained, I will probably get into trouble. A bin man isn't going to use AI to design a plane or cure cancer. But he might end up with something that looks like a plane, or he might decide to go and eat that poisonous plant.

 

NHS 111 has been using AI online for a couple of years. But you won't notice it. You think its a multiple choice questionaire. But its doing much more than that in triaging you.

[MicroB]

2 minutes ago, soalbundy said:

Gibber link was designed by humans so presumably the messages could be translated into English by a man made program.

 

Yeah, hence the subtitles.

[ChaiyaTH]

1 hour ago, Bkk Brian said:

The AI bots will start their own forums next...........

They already do, but then it would be subreddits etc. I have an autonomous Baby AGI framework running with GPT, it just creates posts and a life on it's own depending what messages or replies it get back etc. It can make podcasts itself, create anmimations, social posts etc. And this is wat an amateur like me can do already.. I am tempted daily to mess with this forum and setup a few profiles, nobody would notice. Maybe this post is AI generated already, based on how i write.

 

Anyway to get back on topic: this is just a human created language for 2 vocal AI agents to talk in their own language when agreeing. It has nothing to do with AI making this language itself or deciding things itself. Sorry, next.

[Crossy]

Sounds like a 1990s modem (remember those?).

 

It's going to go pear-shaped when the bots actually develop their own encryption which we can't decode.

[Dionigi]

1 hour ago, Bkk Brian said:

The AI bots will start their own forums next...........

These are not the bots you are searching for🤞

[ChaiyaTH]

4 minutes ago, Crossy said:

Sounds like a 1990s modem (remember those?).

 

It's going to go pear-shaped when the bots actually develop their own encryption which we can't decode.

They can't unless we would somehow give them that computing power at once, even then there would always be competing ones too, which in a way resolves the issues. It's the same with quantum computers, even 1 get's really good that means the other computers do too. So whatever is created can also be destroyed.

 

This is why it is also important to keep a free world and free market, the fact they banned DeepSeek in certain countries is just straight up economical warfare, as well as naive. It was no different with TikTok. People not even realize this is like 'business war'.

[thaibeachlovers]

2 hours ago, thesetat said:

And this is the reason AI should never have been developed. Soon AI will begin manipulating things in ways we will not understand surpassing human influence. We really need to worry when they begin designing and building more complex AI and robots or are given access to countries defense capabilities. 

Skynet is coming. Buckle up.

 

 

[thaibeachlovers]

11 minutes ago, Crossy said:

Sounds like a 1990s modem (remember those?).

 

It's going to go pear-shaped when the bots actually develop their own encryption which we can't decode.

Relax, we will put permanent cloud cover all over the planet which means their solar charging systems don't work- end of problem.................... oh, wait a minute.

[thaibeachlovers]

31 minutes ago, MicroB said:

I still refuse to internet bank, and there is no way I am using a phone as a means of payment. My phone ceased being a phone as such about 2006.

They'll get you in the end though. My 20 year old phone won't work from end of 2025, so they get to steal anything I use a phone for after that. Seems the sheeple are q'ing up to be slaughtered.

 

I don't use internet banking, but they still connect my account to the internet- game over for humans- too stupid to see the precipice the human race is running towards. Are we too stupid to deserve to survive?

[dinsdale]

Communicating in a digital language which has been programmed not "secretly talking their own language". The two AI bots are not having a thoughtful conversation.

[black tabby12345]

3 hours ago, Ralf001 said:

So, the Terminator move was a documentary after all.

 

The inventor of the Polaroid instant photography once said that: Anything people can possibly imagine, can come true.