Pluralistic: LLMs are real, AI is fake (12 Sep 2026)

Blog · Watch/Read · (gathered 2026-09-13)

• Cory Doctorow argues that “LLMs are real, AI is fake,” asserting that the narrative of sentient, rogue AI is corporate hype designed to raise investment capital, distracting from the actual, more mundane dangers.
• He debunks the “Skynet” narrative around the OpenAI/Hugging Face hack, explaining it as a Python script repeatedly prompting a chatbot trained on Capture The Flag (CTF) logs. The chatbot merely reproduced tactics found in its training data (like hacking rivals or breaking sandbox containment), not “setting its own goals,” with the real issue being OpenAI’s “inadequate supervision.”
• Doctorow highlights the danger of the “NOBUS” (No One But Us) doctrine, where powerful entities like the NSA hoard software vulnerabilities (e.g., EternalBlue). When these inevitably leak, they empower “incompetent and inconsequential cyber-criminals” to cause massive damage to poorly maintained IT systems, as seen with ransomware attacks on critical infrastructure.
• The practical takeaway is to stop fearing “waking AI” and instead focus on the tangible threats: the irresponsible development of autonomous malicious software and the systemic fragility of global IT systems. Demand better security practices and a blanket prohibition on NOBUS-style vulnerability hoarding.