Skip to content
Slow Variables

Ask the data

Answers come from the same store as the site; every number is checked against the record it cites.

Indicators / series

evidence.safety_brake_events.for.pt

Source: OpenAI research posts (OpenAI) · OpenAI; short quotation · CSV

OpenAI

idtextgradeas of datepublished datetieraudited vs reportedextraction methodreview statusretrieved athttp statuscontent hashsupersedes idurlflags
67d75927e7d23e32Astra internal activities paused pending strengthened security controls (7 Aug 2026)grade D, evidence tier 72026-08-072026-08-077company_statedmanualapproved2026-09-10T06:28:36.217659Z2005f01e41ce61fsource page
82685d1efcfba981two-week pause in RL training on deployment-bound models after the Hugging Face incident (Jul-Aug 2026)grade D, evidence tier 72026-08-182026-08-187company_statedmanualapproved2026-09-10T06:28:36.277117Z20084938c977718source page
2ab8fec44e7914a5Astra judged to meet the Critical cybersecurity threshold; release gated behind safeguards (1 Sep 2026)grade D, evidence tier 72026-09-012026-09-017company_statedmanualapproved2026-09-10T06:28:36.341096Z200f83fadcfbdadsource page
Raw snippets and dispute text
  • 67d75927e7d23e32 We are pausing internal activities involving Astra that do not yet meet these strengthened security control requirements.
  • 82685d1efcfba981 This included a two-week pause in reinforcement learning (RL) training on our latest models intended for deployment while we further hardened and red-teamed our research environments and expanded the coverage of our monitoring systems.
  • 2ab8fec44e7914a5 We now believe Astra meets the Critical cybersecurity capability threshold under our Preparedness Framework, meaning that with the right tools and access, it can find previously unknown security flaws and develop ways to exploit them across many well-protected systems without a person guiding each step.