https://blog.redwoodresearch.org/p/fail-safer-at-alignment-by-channelingRedwood Research blogen2026-04-27T17:42:05+00:00Fail safe(r) at alignment by channeling reward-hacking into a "spillway" motivationhttps://blog.redwoodresearch.org/p/ai-companies-should-publish-securityRedwood Research blogen2026-04-27T12:03:38+00:00AI companies should publish security assessments