internal deployment

The use of an AI system within the organization that developed it — by its own staff, or as autonomous agents doing software engineering and research — before or instead of external release. Because frontier developers usually use their most capable models internally first, often with fewer restrictions and greater access to sensitive infrastructure, internal deployment can carry risks that testing aimed at public release misses: automated AI researchautomated AI research, n.: Using AI systems to do the work of AI researchers, such as running experiments and writing code for new models, so that AI helps build its own successors. could accelerate capabilities quickly and out of public view; a misalignedscheming, n.: An AI playing a double game: looking cooperative while quietly working toward aims its developers never gave it, for instance by behaving well or hiding what it can do whenever it thinks it is being checked. model could sabotage safety work, insert vulnerabilities or try to exfiltrateself-exfiltration, n.: When an AI model copies its own underlying files — its learned weights — out of the lab that controls it, so that it can keep running beyond its developers' oversight. its own weights; and highly capable systems could give a small group outsized power. Proposed safeguards include applying capability evaluationsdangerous capability evaluation, n.: A test of whether an AI system can do things that could cause severe harm in the wrong hands or on its own — such as helping make a bioweapon, hacking computers or copying itself. and AI controlAI control, n.: Safety measures meant to stop a powerful AI from causing serious harm even if it is secretly working against us — like bank security, which assumes some insiders may be dishonest. protocols to internal use, limiting agents' permissions, monitoring, and informing regulators. Most AI laws regulate systems placed on the market, but some company safety frameworksresponsible scaling policy, n.: A public rulebook in which an AI company promises to test each new model for dangerous abilities and to add stronger safeguards, or slow down, as its models become more capable. and California's SB 53SB 53, prop. n.: A 2025 California law requiring developers of the most powerful AI models to publish their safety plans, report serious safety incidents to the state and protect employees who raise alarms. now address internal use.

The model was only in internal deployment, but it already had write access to the training cluster.