OpenAI Dev Welcomes Prompt-Injection Progress, Calls Astra Most Aligned Model Yet

Tibo ·

Key Info

In a social media response, an OpenAI developer welcomes industry-wide progress on prompt injection and describes Astra as OpenAI's most capable and most aligned model to date. A quoted safety post adds that OpenAI's new model is roughly on par with Gemini Flash and Opus 4.8 on prompt-injection risk, and argues that publicly naming labs encourages them to prioritize safety.

Highlights

  • Prompt injection is framed as an industry-wide problem that is increasingly being solved, with Astra called the most capable and aligned OpenAI model yet.
  • The quoted post says OpenAI's new model matches Gemini Flash and Opus 4.8 on prompt-injection risk.
  • Publicly evaluating and naming other labs is described as an effective way to push them toward more aligned models, with more room for improvement.
  • The quoted post also claims Claude models have had prompt injection practically solved for about two months.
Loading...