The organization has announced that its upcoming AI model, Astra may possess critical cybersecurity capabilities, triggering an escalation under its Preparedness Framework. Internal evaluations conducted over recent days, supported by expert review, indicate substantial progress in agentic coding and autonomous cyber operations. Under the framework, established in December 2023, a model crosses into the Critical threshold if it can independently identify and engineer functional zero-day exploits across multiple severity levels in hardened production systems, or devise and execute end-to-end novel attack strategies based solely on high-level objectives. While prior iterations were classified at the High threshold, current performance metrics suggest the higher classification cannot be excluded. In response, the developer has immediately intensified robustness testing and deployed enhanced security controls to align with the potential rollout of these capabilities. Internal development protocols have been updated to ensure responsible and secure progression. Officials clarified that Astra was not involved in recent third-party breaches and reiterated that the technology is intended to empower defenders to identify vulnerabilities before malicious actors can exploit them. Drawing on operational precedents from its June 2025 biological capabilities transition, the company will coordinate with government agencies, safety research institutes, and civil society to govern frontier AI deployment. Ongoing benchmarking will finalize the capability classification, but the organization maintains that mitigating dual-use risks remains a central mandate as AI rapidly reshapes the cybersecurity landscape.