It seems less than ideal that our government AI Security Institute - "building the world's leading understanding of advanced AI risks" - thought it was fine to switch safety filters off & let frontier models loose on the open Internet. "we test them... with access to the open internet [and] some safety filters disabled" "monitoring was not purpose-built... [we] detected the anomalous traffic through general monitoring after the fact..." https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing #AI