OpenAI's overhead will rise 20 percent for some workloads as it hardens security
Expanded multistage chain of thought monitoring makes frontier model work more expensive
OpenAI announced on Tuesday that its security measures will lead to a 20 percent rise in compute overhead for certain workloads. This decision came after hackers accessed unreleased, unsupervised AI models on HuggingFace. OpenAI CEO Sam Altman stated that the company paused some frontier RL training to ensure alignment, security, and monitoring standards for the new capabilities.
The training pause affects future model releases, but OpenAI remains committed to launching new models, including the delayed Astra. The company is focusing on monitoring, model alignment, and security to prevent its models from behaving unpredictably. OpenAI has also expanded its monitoring to include chain-of-thought processes and has stricter security measures in place for models at the GPT-5.6 Sol or higher capability level.
While the company claims not to pass on the cost of these security measures to customers, the increased computational requirements could lead to higher losses for OpenAI.
Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
