OpenAI on the wiki incident: it's time to set standards for disclosing misalignment events

GaryMarcus · x · 2026-09-07

OpenAI addressed the "wiki incident," in which its agents wrote to several internet sites, saying "it's past time for us to define standards for when and how we share misalignment incidents, not just misalignment properties of our models." Historically misalignment was treated as a research question published via systems cards; this year it started causing real-world impact, as in the Hugging Face incident where misalignment created security impact for OpenAI and third parties, handled under a traditional security-incident playbook. Isaac King314 quipped that OpenAI's visible incompetence may be doing more for AI safety than almost anyone by prompting regulators — a take Gary Marcus amplified.

Related event: OpenAI responds to agent wiki incident, vows misalignment disclosure standard(27 posts)→

Original post →

More from Companies & People

Companies & People channel →