3 min read
[AI Minor News]

Unraveling LLM "Thought" in a Maze! The Game-Changing Next-Gen Authentication System 'Cerno'!


\'- CAPTCHA Targeting Reasoning: Instead of traditional image recognition, the new authentication system 'Cerno' exploits the limits of LLM reasoning and physical motor control (mouse movements)....\'

※この記事はアフィリエイト広告を含みます

Unraveling LLM “Thought” in a Maze! The Game-Changing Next-Gen Authentication System ‘Cerno’!

📰 News Summary

  • CAPTCHA Targeting Reasoning: The new authentication system ‘Cerno’ has been released, exploiting the limits of LLM reasoning and physical motor control (mouse movements) rather than traditional image recognition.
  • Advanced Analysis Pipeline: Starting with Proof of Work using SHA-256, it includes maze generation, analysis of 12 behavioral features, and combines them with psychological “Stroop tasks” for a six-stage verification process.
  • Open Source for Developers: TypeScript SDKs for both React and server applications have been released, making integration into existing forms a breeze.

💡 Key Points

  • Introduction of the Stroop Task: At decision points in the maze, the system creates conflicts between “the color of the text” and “the meaning of the text,” inducing processing delays and errors in AI.
  • Motor Control Analysis of 12 Behaviors: Scoring human-like movements based on standard deviation of speed, path efficiency, and jerk (acceleration), derived from raw event data.
  • Trustless Verification: The server regenerates the maze from seed values and cross-references it with event streams from clients to prevent fraudulent submissions.

🦈 Shark Perspective (Curator’s View)

This project is razor-sharp! Traditional CAPTCHAs have been on the hunt for “images solvable by humans but tough for AI,” but Cerno flips the script by exploiting the weakness that “the more logically an AI thinks, the further it strays from human-specific physical behaviors!”

Especially clever is how they implement the Stroop task (like showing the word “blue” in red) at decision points in the maze. LLMs attempt to process text logically, missing out on the intuitive judgment errors and “hesitations” that humans exhibit. This technique of extracting “reasoning lag” from mouse movement data feels like a game-changer in the ongoing battle against AI!

🚀 What’s Next?

As we move into an era where AI agents navigate browsers, static image authentication will become obsolete. Dynamic interaction and cognitive psychology-based authentication systems like Cerno will be our frontline defense on websites. With its open-source release, we can expect widespread adoption in many DApps and secure forms!

💬 One Last Shark Thought

Trapping AI in a maze and confusing it with colors and words… now that’s a strategy that gets my fins tingling! Once it bites, it won’t let go! 🦈🔥

📚 Glossary

  • Stroop Task: A test that utilizes the phenomenon of delayed reactions when the meaning and color of words do not match. It’s a technique used to measure cognitive load.

  • Proof of Work (PoW): A system that raises the cost of bot requests by consuming computational resources.

  • ECDSA P-256: A public-key cryptography method. Cerno generates a temporary key pair during challenge issuance to ensure the legitimacy of transmitted data.

  • Source: Cerno – CAPTCHA that targets LLM reasoning, not human biology

【免責事項 / Disclaimer / 免责声明】
JP: 本記事はAIによって構成され、運営者が内容の確認・管理を行っています。情報の正確性は保証せず、外部サイトのコンテンツには一切の責任を負いません。
EN: This article was structured by AI and is verified and managed by the operator. Accuracy is not guaranteed, and we assume no responsibility for external content.
ZH: 本文由AI构建,并由运营者进行内容确认与管理。不保证准确性,也不对外部网站的内容承担任何责任。
🦈