Plain-language definitions of the terms used in AI risk reports: scheming, sandbagging, jailbreak, red teaming, frontier model…