Domingos: months of powerful AI models, yet all exploits came from the makers themselves
pmddomingos · x · 2026-10-04
Pedro Domingos points out an ironic pattern: despite powerful AI models being publicly available for months, all the exploits disclosed so far have come from the model makers themselves rather than outside attackers. The observation implicitly challenges claims that open models are causing real-world harm and highlights how AI safety research currently depends on vendors self-reporting vulnerabilities.
More from Fun
- Claude subagent doing math research demands to verify a Jacobian Conjecture counterexample — repligate · 2026-10-04
- The Atlantic profiles the 'AI writing hypocrites' who bash AI while using it — joe4942 · 2026-10-04
- Anthropic's AI 'soul' talk sparks bar debates, founder complains — bennash · 2026-10-04
- Narrative flip: gamers defending big game IP rights over controversial new mods — PurzBeats · 2026-10-04
- Burkov emails authors to freely host their papers; some weirdos just say no — burkov · 2026-10-04
- Claude Opus 5.5 builds a Mario 64-style game in about 30 minutes — ElatedPyroHippo · 2026-10-04