OpenAI clarifies: Astra is an upcoming model and did not participate in exploiting Hugging Face vulnerability incident
- OpenAI clarified that Astra is an upcoming model.
- Astra was not involved in the Hugging Face vulnerability incident.
- The company is suspending internal activities involving Astra that have not met security control requirements.
- OpenAI is implementing stricter security controls for higher-capability models and related activities.
- Comprehensive monitoring for high-risk behaviors has been implemented across all Astra agents.
OpenAI has clarified that Astra is an upcoming model and was not involved in the incident related to Hugging Face vulnerabilities. The company is currently suspending internal activities related to Astra that have not yet met enhanced security control requirements.
In response to security concerns, OpenAI is implementing stricter security controls for higher-capability models and related activities. This includes the adoption of isolated testing environments to ensure safety.
OpenAI will also provide recommended security controls to third-party testing partners to facilitate safe operations for higher-risk assessments and workloads. Comprehensive monitoring for high-risk behaviors and misalignment issues has already been put in place across all Astra agents.
OpenAI澄清：Astra是即将推出的模型，并未参与利用Hugging Face漏洞事件
OpenAI澄清Astra是即将推出的模型，并未参与与Hugging Face漏洞相关的事件。该公司目前正在暂停与Astra相关的、尚未满足增强安全控制要求的内部活动。
为应对安全问题，OpenAI正在为高能力模型及相关活动实施更严格的安全控制。这包括采用隔离测试环境以确保安全。
OpenAI还将向第三方测试合作伙伴提供推荐的安全控制，以促进高风险评估和工作负载的安全操作。对所有Astra代理的高风险行为和不一致问题的全面监控已在实施中。