OpenAI has officially delayed the release of its upcoming 'Astra' AI model, citing the discovery of potential 'critical' cybersecurity capabilities that could be exploited by malicious actors.
The company is currently implementing stricter safety controls and slowing development to ensure the model cannot be used to facilitate cyberattacks or generate harmful code.
This decision follows internal testing that identified the model's advanced ability to identify and exploit software vulnerabilities, prompting a shift in the company's deployment strategy.