Gpt 6 Astra

Technology

• GPT-6 Astra remains vulnerable to hidden prompt injections in 8.5% of cases: OpenAI's GPT-6 Astra blocks 99.99% of direct prompt injections but remains vulnerable to attacks hidden inside documents in 8.5% of scenarios. In comparison, Anthropic's Claude Opus 5 performs better with a 4.8% vulnerability rate. These security gaps pose risks for autonomous AI agents.

  AI Safety • The Decoder

  Permalink