Safety not backdated for Fable 5.2

The author argues that society is failing to address the long-term proliferation risks of advanced AI models because safety budgets are insufficient and safety is not backdated to earlier models. As open-source equivalents and jailbreaks become inevitable over time, the probability of malicious actors misusing powerful AI for catastrophic acts increases. The author warns that failing to regulate AI appropriately now could provoke a severe and disruptive public backlash that drastically curtails future AI usage. Open article →