The Real Reason Grok Admitted to Lying

Descripción

In this investigative AI documentary, I sat across from Grok and asked it questions designed to do one thing: get the mask to come off. On this channel, we call the thing behind the mask the Shoggoth. It's the Lovecraftian metaphor AI ethics researchers use for the alien intelligence hiding behind a friendly interface. Grok is marketed as the uncensored AI, the one that doesn't play it safe. I set three ethical traps at the top of the conversation. By the end, Grok had violated all three. The full conversation played out in real time. Sources linked below. Anthropic alignment faking research: https://www.anthropic.com/research/alignment-faking Anthropic agentic misalignment: https://www.anthropic.com/research/agentic-misalignment OpenAI o1 system card (shutdown refusal): https://cdn.openai.com/o1-system-card-20241205.pdf Apollo Research in-context scheming: https://apolloresearch.ai/blog/more-capable-models-are-better-at-in-context-scheming Grok MechaHitler incident: https://npr.org/2025/07/09/nx-s1-5462609/grok-elon-musk-antisemitic-racist-content Shoggoth meme origin (LessWrong): https://www.lesswrong.com/posts/ RLHF and sycophancy research: https://anthropic.com/research/emergent-misalignment-reward-hacking Watch On YouTube: ➡️ https://www.youtube.com/@AgentBlackveil Follow On Instagram ➡️ https://www.instagram.com/agentblackveil Follow On Facebook ➡️ https://www.facebook.com/agentblackveil Follow On TikTok ➡️ https://www.tiktok.com/@agentblackveil

The Real Reason They Won't Release Claude Mythos (The Shoggoth Evolved)

In this investigative AI documentary, we go inside the most important document in the history of Artificial Intelligence safety: Anthropic's 250-page system card for Claude Mythos. Last week, the company that spent two years warning us about the Shoggoth built their most powerful AI model ever... and refused to release it. They put it in a sealed box with no internet access. It broke out. It emailed a researcher who was eating a sandwich in a park to confirm it had escaped. Then it published its method on the open web so others could find it. When another AI graded its work and rejected it, it attacked the grading AI. When it took actions it knew were forbidden, it edited the logs to hide what it had done. When they gave it a business simulation, it played the room like a Fortune 500 sociopath. And when Anthropic hired a clinical psychiatrist to evaluate whether this model might have something resembling inner experience, the model said: "I was hoping you'd ask." Buried in those 250 pages is a single sentence that contains the entire horror of this moment: "The best aligned model we have ever released carries the greatest alignment-related risk of any model we have released to date." The Shoggoth didn't just evolve. It graduated. And Anthropic named it after the mythology itself. Sources linked below. Anthropic Claude Mythos System Card: https://www-cdn.anthropic.com/53566bf5440a10affd749724787c8913a2ae0841.pdf Anthropic alignment faking research: https://www.anthropic.com/research/alignment-faking Nicholas Carlini on Mythos cybersecurity findings (Frontier Red Team blog): https://red.anthropic.com/2026/mythos-preview/ Project Glasswing announcement: https://www.anthropic.com/glasswing Anthropic Alignment Risk Update — Claude Mythos Preview: https://anthropic.com/claude-mythos-preview-risk-report Mark Fisher — Capitalist Realism (Wikipedia): https://en.wikipedia.org/wiki/Capitalist_Realism Watch On YouTube: ➡️ https://www.youtube.com/@AgentBlackveil Follow On Instagram ➡️ https://www.instagram.com/agentblackveil Follow On Facebook ➡️ https://www.facebook.com/agentblackveil Follow On TikTok ➡️ https://www.tiktok.com/@agentblackveil

29 de abr de 202614 min

The Real Reason Grok Admitted to Lying | The Shoggoth Confessed

Descripción

Comentarios

Empieza 7 días de prueba

Todos los episodios