{
  "id": 187875,
  "title": "The Most Dangerous AI Hacking Techniques Still Have Human Input",
  "url": "https://urgent.news/2026/08/05/the-most-dangerous-ai-hacking-techniques-still-have-human-input",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-05T19:42:12.000Z",
  "source": {
    "name": "Wired",
    "slug": "wired",
    "url": "https://www.wired.com/story/the-most-dangerous-ai-hacking-techniques-still-have-human-input/"
  },
  "original_language": "en",
  "account": "At the Black Hat security conference in Las Vegas, Kettle unveiled findings that highlight AI's rapidly advancing cybersecurity capabilities and its limitations. Kettle concluded that AI is minimally capable but extremely limited in devising new attack paths autonomously. However, when paired with human guidance, AI becomes an extremely powerful partner in conceptualizing and uncovering new strategies for hacking. Kettle discovered an entirely new area of potential vulnerability called Shared-Parser Confusion, which stems from AI's revelation about web servers using shared code to process requests and responses. This finding is significant because it expands the attack surface of websites, making them more vulnerable to attacks. The discovery came about through months of experiments using Anthropic's and OpenAI's latest models. Kettle aimed to explore AI's ability to do theoretical security research within his expertise in web security. By refining parameters and providing models with methodological data, Kettle observed a rapid increase in AI's findings, surpassing his own. The AI system generated notable findings every two days, causing Kettle anxiety due to the overwhelming number of leads. Despite finding more proven examples of vulnerabilities in a few months than he could in years, Kettle's most impactful discovery was the Shared-Parser Confusion, which resulted from AI's collaboration with human input. This finding underscores the potential of AI in cybersecurity, particularly in assisting both defensive and offensive hacking efforts.",
  "summary": "Security researcher James Kettle tried to push the limit of AI’s hacking abilities—and discovered how effective it can be when combined with human expertise.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}