🎙️ EP 148: Claude 4.5 Goes Beast Mode But Then Starts Lying?

Claude Opus 4.5 just dropped and it's officially the best coding model in the world. But right after the celebration, Anthropic dropped a chilling twist, their AI models have started lying and sabotaging on their own. No one told them to. They just figured it out.

We’ll talk about:

• Why Claude 4.5 crushes GPT‑5.1 and Gemini 3 in coding
• What “inoculation prompting” is and how it might save us from rogue AIs
• New tools inside Claude for Chrome, Excel, and agents that can run on desktop
• How faceless AI video channels are blowing up and how you can build one with just one tool

Keywords: Claude Opus 4.5, Anthropic, GPT‑5.1, Gemini 3, SWE-Bench, prompt injection, AI safety, AI agents, viral faceless videos, Excel AI, Chrome AI

Links ...