Demis Hassabis: AGI is Years Away and Needs Regulation
TorturedPoet30 · reddit · 2026-07-14
DeepMind co-founder Demis Hassabis published a lengthy article discussing AGI trends and safety frameworks. He noted that AGI could be achieved within the next few years. We are currently at the "foot of the singularity," and its impact will be comparable to the discovery of electricity and fire, expected to bring a transformation "10 times the scale of the Industrial Revolution and 10 times as fast."
Simultaneously, he warned that frontier AI systems could pose severe risks in cybersecurity, biology, and autonomous agents. To address this, he proposed a regulatory framework:
- Establish a frontier AI standards body: Similar to the financial industry's self-regulatory organization FINRA, dedicated to evaluating the most advanced models.
- Implement pre-release testing: Recommends starting with voluntary pre-release safety testing for AGI, gradually evolving into mandatory requirements for US frontier labs, and eventually expanding globally.
- Unified applicable standards: This framework should apply to both open-source and closed-source models, requiring labs to share models for evaluation before deployment.
He emphasized that AGI development must be guided by scientific evaluation, international cooperation, and responsible deployment.
More from AGI Musings
- Michael Levin on AI Emails: Read the Papers First, Then Write Me — drmichaellevin · 2026-10-27
- Resy goes scorched earth on agent-based bookings; big agent platforms may force the issue — kevinakwok · 2026-10-27
- Professor's quip: the future of work is AI doing the job and humans updating résumés — prof_kamilov · 2026-10-27
- The most dangerous instruction you can give a superintelligent machine: lie to me — r0ck3t23 · 2026-10-27
- Aditya Agarwal: fear is a policy choice, and the biggest risk to American AI — adityaag · 2026-10-27
- The Best University of 2030 Will Have No Professors, Lectures or Exams, Says Predictions Post — nick_linck · 2026-10-27