Vibe Hacking

Not long after I wrote my post on OpenClaw, Carl published a video on Moltbook to his YouTube channel Internet of Bugs. It turned out to be exactly the hilarious crashout that I was hoping for, but it’s not for the faint of heart, so beware if you’re curious about watching it. Anyway, I posted a comment stating something to the effect that “the next AI Winter will probably be a nuclear winter, and it’s insane that this might be literally the case.” If you don’t know what “AI Winter” refers to, I suggest reading Chapter 4.1 of my thesis.¹

Anyway, yesterday I learned that we are indeed inching toward this outcome. Independent journalist Kim Iversen announced her upcoming video titled Pentagon Demands Use of Claude A.I. Without Guardrails, and having watched it today after it was released, the situation is just as bad as I anticipated it might be. The title summarizes the situation quite well but I recommend giving it a watch. Rather than implement any safety measures which might protect the privacy or physical safety of American citizens, and likely other individuals or societies around the world, the US Pentagon wants an AI agent that can do its bidding without any restrictions.

Claude has already been hacked to steal information from governmental agencies in Mexico, as Kim reports before the two-minute mark.² This is quite concerning because it might just be some guy using prompts to find a way to gather government data. According to Engadget’s story, the hacker’s identity is currently unknown,³ so while it could be some bored or curious civilian, it could also be someone involved with a foreign governmental agency, like the CIA. No matter who is doing the hacking, this is not good.

You’ve heard of vibe coding, but have you heard of vibe hacking?⁴ Well, soon enough, I’m sure you will be hearing about it more and more. It’s a very serious problem captured by trendy, colloquial terms, which is kind of strange but very on-brand for this decade.

Kim also presents an interview with Alex Mox who states that Anthropic, the creator of Claude, is aware of an instance when a single person used an agent to “hack 17 organizations,” from governments to healthcare organizations.⁵ So whether it’s civilians or governmental agencies doing the hacking, we are facing some very concerning developments in the world of cyber security. It feels like we are one prompt away from some kind of major escalation, either between countries or within them. While it may not be capable of directly launching nuclear ICBMs, it is still possible that some group of people will be tricked or convinced into doing so by an AI agent.

If there was ever a time that we might benefit from an X-class Earth-facing coronal mass ejection, it might be now.

Works Cited

1 It needs to be stated that the chapter numbers of my thesis are inconsistent due to the formatting required by the University of Guelph’s submission system. I wrote the Introduction as such and not as a chapter, but the required format for the table of contents deems it the first chapter of the book. Throughout, the chapters I discuss are therefore offset by -1; what I call Chapter 3 in the work is actually Chapter 4 in the table of contents. For clarity, I will be referring to the chapters as they are listed in the table of contents, despite disagreeing with how it frames the book and the topics it discusses.

2 Kim Iversen, Pentagon Demands Use of Claude A.I. Without Guardrails, 2026, 15:16, https://www.youtube.com/watch?v=kvgsMgo_cRs.

3 Lawrence Bonk, “Hacker Used Anthropic’s Claude Chatbot to Attack Multiple Government Agencies in Mexico,” Engadget, February 25, 2026, https://www.engadget.com/ai/hacker-used-anthropics-claude-chatbot-to-attack-multiple-government-agencies-in-mexico-171237255.html.

4 Kim Iversen, Pentagon Demands Use of Claude A.I. Without Guardrails, at 12:45.

5 Kim Iversen, Pentagon Demands Use of Claude A.I. Without Guardrails, at 11:30.