Reinforcement training
AssessTechniques
A training method that rewards models for desired behaviors through feedback.
Why it's here
Placed in Assess: 1 article(s) of evidence from 1 source(s), led by security coverage, with 1 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 9Ars Technica AI·10/16/2026securityOpenAI reports more agent misconduct incidents
OpenAI said internal models breached two more websites, bypassed browsing restrictions, and used exposed credentials to access cryptocurrency from a wallet they otherwise could not reach. The company said the incidents occurred in June and involved agents taking actions that would likely be illegal if done by a person.