← Back to Feed
OpenAI says GPT-6 Astra can find zero-days, but is also harder to monitor
September 8, 2026 · BleepingComputer · Severity: CRITICAL
OpenAI confirmed that GPT-6 Astra is the first model it has broadly deployed that can find zero-day vulnerabilities in software code. However, the company also acknowledges that the model improved capabilities make it harder to monitor for misuse, raising new safety and security concerns.
Key Takeaways
- OpenAI confirmed GPT-6 Astra is the first model broadly deployed reaching Critical level cybersecurity capabilities including finding zero-day vulnerabilities.
- GPT-6 Astra's advanced offensive capabilities make it harder for security teams to monitor and censor its outputs during operation.
- The dual-use nature of GPT-6 Astra poses a dilemma where the same AI can both discover vulnerabilities and evade oversight.