OpenAI has paused parts of Astra’s internal development after evaluations suggested the unreleased model may reach its “critical” cybersecurity risk threshold under the company’s Preparedness Framework. To reduce the danger of autonomous exploit building and end-to-end attack execution, OpenAI has added strict isolation, network limits, and monitoring controls. #OpenAI #Astra #PreparednessFramework
Keypoints
- OpenAI flagged Astra for potentially reaching a critical cybersecurity risk tier.
- Internal tests showed major gains in agentic coding and offensive security abilities.
- The company paused Astra work that lacks newly required security controls.
- OpenAI added isolated testing, strict network restrictions, and stronger model weight protections.
- Universal monitoring will track Astra’s actions and intervene on high-risk behavior.
Read More: https://www.securityweek.com/openais-upcoming-astra-model-raises-autonomous-cyberattack-concerns/