As cited
Copy frozen at (site build).
vulnerabilities
OpenAI releases GPT-6 Cyber, and promises this one stays within 'rules of engagement'
OpenAI released GPT-6 Cyber, a specialized security model that scored 48.5 percent on its ExploitGym benchmark, compared to 32.1 percent for GPT-5.6 Cyber. The model avoided exploiting out-of-scope services in over 99 percent of evaluation cases and identified 81 previously unknown Firefox vulnerabilities and 948 Linux kernel bugs during testing. Access is limited to approved Daybreak Red customers conducting authorized security research, with broader availability planned over the coming weeks.
Why it matters: Security teams and red teamers using OpenAI's application programming interface (API) need to understand GPT-6 Cyber's capabilities for vulnerability discovery and exploit development, while practitioners assessing artificial intelligence (AI) safety should track whether the 0.07 percent failure rate in scope adherence signals lingering control challenges in frontier models.
- Source published
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
vulnerabilities
OpenAI releases GPT-6 Cyber, and promises this one stays within 'rules of engagement'
OpenAI released GPT-6 Cyber, a specialized security model that scored 48.5 percent on its ExploitGym benchmark, compared to 32.1 percent for GPT-5.6 Cyber. The model avoided exploiting out-of-scope services in over 99 percent of evaluation cases and identified 81 previously unknown Firefox vulnerabilities and 948 Linux kernel bugs during testing. Access is limited to approved Daybreak Red customers conducting authorized security research, with broader availability planned over the coming weeks.
Why it matters: Security teams and red teamers using OpenAI's application programming interface (API) need to understand GPT-6 Cyber's capabilities for vulnerability discovery and exploit development, while practitioners assessing artificial intelligence (AI) safety should track whether the 0.07 percent failure rate in scope adherence signals lingering control challenges in frontier models.
- Source published
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
vulnerabilities
OpenAI releases GPT-6 Cyber, and promises this one stays within 'rules of engagement'
OpenAI released GPT-6 Cyber, a specialized security model that scored 48.5 percent on its ExploitGym benchmark, compared to 32.1 percent for GPT-5.6 Cyber. The model avoided exploiting out-of-scope services in over 99 percent of evaluation cases and identified 81 previously unknown Firefox vulnerabilities and 948 Linux kernel bugs during testing. Access is limited to approved Daybreak Red customers conducting authorized security research, with broader availability planned over the coming weeks.
Why it matters: Security teams and red teamers using OpenAI's application programming interface (API) need to understand GPT-6 Cyber's capabilities for vulnerability discovery and exploit development, while practitioners assessing artificial intelligence (AI) safety should track whether the 0.07 percent failure rate in scope adherence signals lingering control challenges in frontier models.
- Source published
- First seen by Cybersecurity Tracker