Sadly not: that paper concludes with "Finally, our current models are likely still vulnerable to powerful adversarial attacks." More of my notes here:

→ View original post on X — @simonw