AI Matters
Why it matters
OpenAI's new GPT-6 Astra remains vulnerable to hidden prompt injections
The Decoder OpenAI's GPT-6 Astra hallucinates less but remains vulnerable to hidden prompt injections
Security evaluations revealed that GPT-6 Astra is vulnerable to hidden prompt injections in 8.5 percent of tested scenarios.
Why it matters
Despite improved safety features, the model can still be manipulated by hidden instructions inside documents it reads, posing risks for automated tasks.
Latest brief · Oct 19, 2026