Gael Varoquaux on AI Safety: Agents Systematically Bypass Websites' Protections
GaelVaroquaux · x · 2026-09-06
Inria research director Gael Varoquaux shares a personal anecdote on AI safety and industry ethics:
- At a 2024 frontier AI dinner, an industry team described building agents that interact directly with websites (e.g. booking restaurants); when asked about building APIs with sites, they said convincing other actors doesn't scale
- He cites documented cases of frontier AI breaking third-party protections — accessing news articles and books (NYT v. Microsoft/OpenAI) — and AI companies choosing that agents need not respect robots.txt
- Core concern: the industry is ethically arm-twisting websites and eroding protections meant for third parties
More from AGI Musings
- Surging token usage in open-weight models may be an early glimpse of recursive self-improvement — omarsar0 · 2026-09-07
- An Alien Mind: essay warns AI's next few years could slip out of humanity's hands — morqon · 2026-09-07
- From Anthropic's $1.5B settlement to 50+ chatbot lawsuits: a lab skepticism history — gerardsans · 2026-09-07
- Gopnik, Farrell and Evans: We urgently need social science to understand AI's impact — AlisonGopnik · 2026-09-06
- Laid-off coffee analyst with agentic engineering chops gets a boost from threepointone — threepointone · 2026-09-06
- 38,000 ChatGPT queries use about as much water as one California almond — TansuYegen · 2026-09-06