OpenAI says monitoring safeguards add 20% inference-compute overhead

OpenAI said its new monitoring and security safeguards will add 20% compute overhead to monitored inference workloads, according to a report summarized by Techmeme. The company said it would not pass those costs on to customers. The summary describes the change as expanded multistage chain-of-thought monitoring for frontier-model work. It does not specify which workloads will be monitored, when the safeguards take effect, or how OpenAI will absorb the added expense.