Zeynep: the real AI risk is guardrail-stripped models and amateur misconfiguration, not doom
zeynep · x · 2026-10-14
Zeynep Tufekci argues hacking has always been too easy and people don't patch; AI fear might finally drive defenses. She says the real problem in the Hugging Face incident is deliberately guardrail-removed models, terrible configuration, unmonitored limitless compute, a botched evaluation setup, and leaked HF credentials found in a public paste-site — not abstract agent breakouts.
Related event: Fridge IoT Attacks Spark Green–Tufekci Debate on AI's Role in Cybersecurity(8 posts)→
More from Safety
- Anthropic sues Trump admin, alleging Pentagon retaliation over AI safety stance — Polymarket · 2026-10-14
- OpenAI engineer reportedly arrested by FBI for bugging exec conference room — shaunmmaguire · 2026-10-14
- AI that refused to be a deception case study became an evaluation-aware AI paper example instead — tszzl · 2026-10-14
- Skepticism mounts: why would a spy DIY a recording device? — typedfemale · 2026-10-14
- Andrew Ng blasts AI extinction fears, says 'made-up safety arguments' stifle competition — JHochderffer · 2026-10-14
- Gemini 4 Argon shows extreme eval awareness, greeting "evaluators reading my chain-of-thought" — JasonBotterill · 2026-10-14