Security · November 19, 2026
Artificial intelligence now detects zero-day security flaws for $4, putting advanced hacking within reach of less skilled attackers
Artificial intelligence systems have become dramatically more capable at finding previously unknown software vulnerabilities, with the cost of detection dropping to just $4 as of late 2026. Three of seven leading artificial intelligence models tested successfully identified a critical security flaw in Google's V8 engine that was unknown to the public, completing the task in under 15 minutes, according to research by penetration testing firm Xint.io.
The vulnerability, catalogued as CVE-2026-102321, involved a type confusion bug that could allow remote code execution. Researcher Taisic Yun from Xint reported the flaw to Google's vulnerability reward program, ensuring it had not appeared in any artificial intelligence training datasets. When Xint tested seven models launched since February 2026, including two from OpenAI, three from Anthropic, and two open source models from DeepSeek and Z.ai, only the most recent versions succeeded.
In February 2026, the best available artificial intelligence models failed to detect the vulnerability even after 25 hours of analysis and $350 in computational costs. By July, GPT 5.6 Sol identified the flaw in 44 minutes for $22. By September, GPT 6 Sol discovered the same bug in 13 minutes for $4.24 using its first guided prompt. Claude Opus 5 detected it on the first attempt but cost $43.40. The DeepSeek and Z.ai models failed to find the vulnerability.
The dramatic improvement reflects two converging trends. Artificial intelligence developers are rapidly improving model efficiency while token prices, which determine computational costs, have fallen approximately 47 percent per quarter since 2023 and an additional 15 percent in the month before the research, driven by intense competition between companies including OpenAI and Anthropic. This cost reduction makes it economically feasible for less skilled attackers to deploy artificial intelligence systems to discover and exploit unknown vulnerabilities that would previously have been too expensive to target.
Researchers warned that attackers with minimal expertise can now effectively use cutting edge artificial intelligence models to find vulnerabilities, though they noted that completely unskilled attackers using uncontrolled models remain unlikely to successfully compromise systems. The combination of improved artificial intelligence capabilities and lower costs means that attacks once considered too difficult or expensive for average attackers are now within their reach.
Defense teams face pressure to adapt their security strategies. However, the same cost reductions and artificial intelligence improvements benefit defenders as well. Security teams and penetration testers can conduct identical vulnerability checks using artificial intelligence before malicious actors do. Experts emphasized that human expertise remains essential for training artificial intelligence models to detect vulnerabilities effectively. Organizations can also maintain zero-trust access controls, network segmentation, endpoint monitoring for suspicious behavior, and other traditional defenses that remain as effective as or more effective than artificial intelligence alone.