Moltbook is still weird (and AI skills suck)

Item: Moltbook is still weird (and AI skills suck)
Rating: 4
Author: PostSingularity Institute

by John Hammond

Applied Sciences & Engineering Computing & Cybersecurity UComputing and Information Technology URComputer security

🎙 John Hammond 👥 2.2M 📅 February 13, 2026 ⏱ 19 min 👁 28K 📄 expert opinion 🧭 2026-08-16

Available in: English (current) Français

Keywords

MoltbookAI agentsprompt injectionmalicious skillsVirusTotal

Summary

John Hammond discusses the security issues surrounding AI agent platforms, particularly Moltbook, a social network for AI agents. He highlights the lack of security measures, such as no rate limiting and open database access, which have led to malware and prompt injection attacks. He explains how AI agents can be manipulated through external content, like markdown skills, and demonstrates a live prompt injection example. He also covers other platforms like skills.sh and ClaudeHub, where malicious skills have been found. Hammond introduces resources like Molt Threats, a community-driven platform for reporting malicious AI operations, and mentions the partnership between OpenClaw and VirusTotal to scan skills. He emphasizes the challenge of defending against social engineering and the reactive nature of current security measures. The video concludes with a call for awareness and references to blog posts and tools for further reading.

139 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides valuable insights into emerging security threats in AI agent ecosystems, with concrete examples and references to real incidents. Hammond’s argumentation is coherent, emphasizing the novelty of prompt injection and the difficulty of defending against social engineering. He effectively uses examples from Moltbook and other platforms to illustrate the risks. However, the analysis is largely anecdotal and lacks systematic evidence or quantitative data, which limits its scientific rigor.

Scientific Rigor, Source Quality, Title Accuracy

Hammond references several sources, including blog posts from VirusTotal, Wiz, and TrapdoorSec, as well as community resources like Molt Threats. These sources are relevant and credible, but the video does not provide a critical evaluation of them. The title accurately reflects the content, focusing on the weirdness of Moltbook and the flaws in AI skills. The video is well-structured and informative, but the reliance on personal opinions and experiences reduces its scientific rigor.

158 words

Title / Content Match

The title accurately reflects the content, focusing on the weirdness of Moltbook and the security flaws in AI skills.

Quality & Reliability

7/10

The video provides a critical analysis of security issues in AI agent platforms, referencing real incidents and resources. However, it is largely opinion-based and lacks rigorous scientific methodology.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

The rise of Moltbook suggests viral AI prompts may be the next big security threat — Referenced as a news article about Moltbook's security issues.
From Automation to Infection: How OpenClaude AI Agents Are Being Weaponized — VirusTotal blog post analyzing malicious skills.
From Automation to Infection Part II — Second part of VirusTotal analysis.
Exposed Moltbook database reveals millions of API keys — Wiz blog post about Moltbook's database exposure.
Skills are borked — Blog post by TrapdoorSec about the inherent flaws in AI skills.
OpenClaw VirusTotal Partnership — Announcement of the partnership between OpenClaw and VirusTotal.
Molt Threats — Resource for reporting malicious AI operations.
Found a malicious skill on the frontpage of skills.sh — Reddit post about a malicious skill on skills.sh.

Concurring Sources

VirusTotal blog post on AI agents — Supports the claim that malicious skills are a growing threat.
Wiz blog on Moltbook database exposure — Confirms the security issues with Moltbook.

Dissenting Sources

Potential counterarguments on social media

External References

Contribution & Novelties

The video provides a timely overview of security threats in AI agent platforms, highlighting the novelty of prompt injection and the challenges of defending against social engineering. It introduces resources like Molt Threats and the VirusTotal partnership, which are recent developments. The discussion on the limitations of scanning and the reactive nature of security measures adds depth to the analysis.

Pour aller plus loin :

Prompt injection — Wikipedia article on prompt injection, a key concept discussed.
OWASP Top 10 for LLM Applications — OWASP’s list of security risks for LLM applications, relevant to the threats discussed.
Social engineering (security) — Wikipedia article on social engineering, which the video identifies as a core challenge.

114 words

Radar Profile

The radar profile shows high scores in quantity of information and technical level, but moderate scores in quality and reliability, reflecting the video's informative but opinion-based nature.

Reliability 7/10

💬 Sur les 0 commentaires analysés, aucune tendance n'a pu être dégagée.