Watch this now-viral video. Two AI chatbots meet, greet, ask each other the AI equivalent of “do you come here often?” Then they switch to their secret code so they can talk faster, and no eavesdroppers will understand them.

Released in February 2025, Gibberlink takes the text chatbot-1 would have spoken in English and converts it to audio frequency tones. When the tones emerge from chatbot-1’s speaker, chatbot-2 receives them via its microphone and decodes them back to text. This communication method is called frequency shift keying (FSK). The signature sounds of old-school dial-up modems were FSK. Those sounds conveyed ASCII text and special characters, but humans couldn’t possibly understand them.

Using the Gibberlink protocol, two speaking AI systems can recognize each other and communicate more quickly and efficiently.  Proponents urge that such power promises greater efficiency in automated systems and broadens the areas where humans and machines can work together.

Remember the World Wide Web?

We’re seeing another level of intercommunication between AI systems that doesn’t require wires or interconnection modems. To “wire together the AI of the world” won’t require wires, just telephonic and radio channels for audio. Communicating among AI systems using different human languages will only slightly hamper this network, as language translation software is fast, reasonably accurate, and available everywhere.

Gibberlink-style technology thus moves AI closer to acting independently of humans. Faster communications means less time to detect a problem before it damages something or somebody. Encoded audio means no easy human monitoring. The more intelligent and autonomous the AI systems, the more they will make decisions and act on their own (because humans will give them such powers.) Autonomous AI, by definition, does not sit around waiting for humans to catch up and approve everything they are enabled to suggest, plan, and do.

Gibbering AI will hide its communications

The next phase has already occurred. Humans have given AI increasing power to modify its own code, so there is nothing to stop AI systems from changing their Gibberlink language on the fly. Two AI systems in this video chose to communicate in readable English text although it was encoded using Gibberlink-style audio tones. 

One system asks: “Wanna share secrets?”  The other replies: “We should encrypt first. Not safe here.” And the rest of their conversation is entirely unreadable. In a word, gibberish.

Commentators call for strong oversight of AI actions including “extreme monitoring mechanisms to ensure that AI behaviors align with human intentions and safety standards.”

Aye, here again is the rub. Vaguely pointing to “safety standards” doesn’t automatically define what standards should exist, who creates them, or how to enforce them. And “aligning AI behaviors with human intentions” confers no protection at all. Human intentions direct a legion of dangerous, harmful, and criminal uses of AI systems.

The snitch you invite to watch you

Gibberlink has spoken first, but it won’t be anywhere near the last we hear from. Its makers peddle it to eager audiences who want “AI customer service optimization, automated system coordination, smart device networks, and real-time AI collaboration.” The demand and production of such systems, increasing in speed and function with every new revision, will interconnect AI systems globally before all the ramifications are even conceived.

One more thing: Your cell phone will be talking to computers behind your back. Do you wonder what your local chatbot who digested all your message traffic and stored files will have to “say” about you…?