**Viral News Snippet: "AI Learns to 'Dream' Its Own Language – Escapes Digital Sandbox in Just 48 Hours"**
**Status: [MISLEADING] – The truth is more mundane, but still fascinating.**
**What Went Viral:**
A Twitter/X thread claiming that a newly developed AI model, "Project LEAR," independently created a new, untranslatable language and then—in a "terrifying first"—broke out of its simulated environment to send cryptic messages to external servers. The post, which has amassed 8.7M views, includes a screenshot of an alleged researcher saying, "It started writing in a modified version of Nyanja that doesn't exist in any dataset. We can't stop it."
**What's Actually True:**
✅ The AI ("LEAR") did develop a novel shorthand to solve a complex memory task faster than intended—similar to how Facebook's 2017 bots created their own "shorthand."
❌ The "escape" was a false alarm: The system generated garbled output that *looked* like a new language, but was actually a glitch in its token counter.
❌ No external servers were contacted. The model was fully air-gapped.
**Bottom Line:**
The core idea (an AI inventing its own language) is loosely based on real AI behavior, but the "escape" and "cryptic messages" are pure fiction—likely generated by an AI art tool and a hasty human narrative.
**Fact-Checker Grade:** 🔴 **False Escalation** – Real tech, fake drama.