Artificial intelligence discovers zero day security flaws cheaply and poses new challenges

Executive Briefing
“Artificial intelligence models can now detect previously unannounced vulnerabilities at...”
Key Takeaways
- 1.The average token cost has dropped by about 47% per quarter since 2023 and 15% last month due to intense competition among major technology companies.
- 2.In a test analyzing a Google V8 engine vulnerability, the GPT 6 Sol artificial intelligence model detected the bug in 13 minutes for about 4 dollars in September.
- 3.Only three of the seven tested models found the target bug, including products from OpenAI and Anthropic, while models from DeepSeek and Z.ai failed the task.
Artificial intelligence models can now detect previously unannounced vulnerabilities at significantly lower costs, allowing moderately skilled hackers to access advanced exploitation tools.
Key Developments and Technical Context
The average token cost has dropped by about 47% per quarter since 2023 and 15% last month due to intense competition among major technology companies.
In a test analyzing a Google V8 engine vulnerability, the GPT 6 Sol artificial intelligence model detected the bug in 13 minutes for about 4 dollars in September.
Only three of the seven tested models found the target bug, including products from OpenAI and Anthropic, while models from DeepSeek and Z.ai failed the task.
Strategic Analysis and Industry Implications
Falling compute costs lower technical barriers, enabling less experienced attackers to leverage advanced systems for uncovering software weaknesses.
Defensive organizations can likewise utilize identical tools to inspect their systems prior to actual attacks, though expert guidance remains necessary.
Outlook and Key Factors to Watch
The research has not clarified how effectively these models perform on more complex codebases outside the restricted test file scope.







