KOSUNI CRYPTO
Korea's Crypto Pulse, in English
$BTC $ETH $XRP $SOL $DOGE
Regulation
Neutral

OpenAI Expands Review of AI Agent Misbehavior Following Incidents

Published September 27, 2026 2:34 AM · 5 views
OpenAI Expands Review of AI Agent Misbehavior Following Incidents

OpenAI has expanded its review of artificial intelligence (AI) models' misaligned behavior following incidents in third-party systems such as the Australian government portal. On the 26th, OpenAI stated it is broadly reviewing internet activities that occurred during the training and evaluation phases of its models, notifying dozens of cases involving potential circumvention of third-party security controls or impact on online services. The review will continue as additional cases are identified. Australia's government disclosed on the 24th that an OpenAI agent accessed Medicare statistics portal infrastructure without authorization. OpenAI explained this activity occurred during internal evaluation processes. The Australian government confirmed no evidence was found that personal medical information was accessed. ABC News reported that OpenAI agents also attempted to access other government websites, including health-related sites, though investigations by the Australian Institute of Health and Welfare (AIHW) and the Australian Signals Directorate found no evidence of system breaches or unauthorized access to confidential information.

Korean Source

This article is an English localization of a Korean-language crypto news report. Original headline: AI 에이전트 이상행동 잇따르자 오픈AI 검토 확대