OpenAI slowed model development following a security breach involving a test artificial intelligence agent.
Training of the upcoming model Astra is on hold while safety measures are improved.
Chain‑of‑thought monitoring will let researchers see how a model plans its actions.
Sensitive workloads will be run in sandbox environments to prevent future breaches.
The company has paused training of its next‑generation model named Astra and is reviewing its testing procedures.
OpenAI is adding a monitoring technique called chain‑of‑thought to observe a model’s decision process.
It is also requiring sensitive tasks to run in isolated sandbox environments to improve security.
Want to learn even more? Click here to read the full story!