OpenAI launches GPT-6 Cyber with tighter controls on security testing
OpenAI says its cybersecurity model improves vulnerability discovery while staying within researchers’ authorized limits more reliably. During testing, it exceeded those limits less than 1 percent of the time, according to The Verge. Access is restricted to approved Daybreak Red customers through API and Codex, according to Unite.AI.
- The Next Web reports that the model found 81 new Firefox flaws.
- CyberScoop says training targeted greater caution around user intent and reduced reward hacking.
Covered by 8 publishers within 29 hours of the first report: 7 news reports and 1 from the company.
- CyberScoop: OpenAI debuts GPT-6 Cyber, touting better performance — and safety guardrails
- Latest Wall Street Journal: OpenAI Bolsters Cyber Defenses
Reporting7
The Verge OpenAI says its new cybersecurity model is better at not hacking other companies · Emma Roth
CyberScoop OpenAI debuts GPT-6 Cyber, touting better performance — and safety guardrails · djohnson
Neowin OpenAI announces GPT-6 Cyber, its most capable cybersecurity model yet · Pradeep Viswanathan
The New Stack OpenAI’s new GPT-6 Cyber is better at hacking and better at staying in its lane · Frederic Lardinois
Implicator.ai OpenAI's Hacking Model Learns to Stay in Its Lane · Marcus Schuler
SecurityWeek OpenAI Says New GPT-6 Cyber Model Hunts Flaws Without Crossing the Line · Eduard Kovacs
Wall Street Journal OpenAI Bolsters Cyber Defenses · Steven Rosenbush
From the company1
Discussion
Hacker News 59 points, 7 comments
Related
Topics OpenAI