OpenAI Warns New Model Could Pose Security Risks

OpenAI tested its new Astra model and found it could perform complex coding and cybersecurity tasks. The company says this performance reaches a dangerous level of capability that it previously hoped to avoid. This is the first time they have officially admitted a model might cross this safety line.

Because of these findings, the company is taking immediate action to manage the risk. They have applied stricter security controls and temporarily paused some development work on the model. These steps are designed to prevent the software from being misused before it is released.

OpenAI plans to bring in outside experts to evaluate the situation further. They want to ensure the technology is safe before moving forward. This development highlights the growing challenges companies face as they build more powerful artificial intelligence.