Inside the Rise of Autonomous AI Hackers: XBOW's Oege De Moor

Why It Matters

Autonomous AI hackers can breach systems faster than vulnerabilities are disclosed, forcing the industry into an urgent AI‑driven defensive arms race.

Key Takeaways

• Autonomous AI hacker Xbo outperformed human hackers on HackerOne
• Xbo discovered remote code execution bug in Bing image search for $3k
• AI‑driven attacks now exploit vulnerabilities before CVEs are published
• Model “alloys” combining Gemini and Sonnet outperform single LLMs
• Industry must accelerate defensive AI tools within six‑to‑nine months

Summary

The presentation highlighted the emergence of fully autonomous AI hackers, focusing on Xbo, a system built by XBOW that can locate, exploit, and report vulnerabilities without human input. Xbo’s most notable achievement was discovering a remote code execution flaw in Microsoft’s Bing image search for a modest $3,000 cost, then climbing to the top of HackerOne’s global leaderboard by conducting black‑box testing solely from a URL. Key insights included the speed at which AI‑driven attacks now outpace traditional vulnerability disclosure: many CVEs are exploited before they are publicly reported. Xbo leverages a hybrid “alloy” of Gemini and Sonnet models, a pair‑programming approach that mitigates individual model errors and delivers performance three times better than the best human hackers, with future GPT‑5‑level models expected to be even more potent. The speaker invoked historical analogies, likening today’s AI arms race to Oda Nobunaga’s gun‑armed forces defeating the cavalry‑centric Takeda clan. He warned that the lag between CVE publication and exploitation has turned negative, and urged defenders to adopt AI tools now, noting that without rapid action, even a Thanksgiving dinner could be jeopardized. Implications are clear: cybersecurity firms must treat AI as both a weapon and a shield, accelerating defensive model development within the next six to nine months. Failure to do so risks a market shift where traditional security stocks tumble as autonomous AI attackers dominate the threat landscape.

Original Description

Oege de Moor, founder and CEO of XBOW, takes the AI Ascent 2026 stage to argue that autonomous AI hackers are already winning. He explains how XBOW's bot became the #1 ranked hacker on HackerOne in August 2025 using only black-box access, how it found a vulnerability in Bing Image Search at a list price of $3,000, and how GPT-5 would have made the same system three times more effective. He closes with a call to action: frontier labs need to maximize the cyber capabilities of their models, defenders need to start using AI offensively to find their own vulnerabilities, and we have roughly six to nine months before open-weight models reach the same capability, at which point everyone, including bad actors, gets the same tools.

00:00 Autonomous Hacking Threat

00:37 Cybersecurity Arms Race

01:34 Bing RCE Case Study

02:32 How ExBo Attacks

03:05 Proving It on HackerOne

04:19 Model Alloys Explained

04:45 Scaling With New Models

05:19 Mythos vs Real Exploits

06:28 CVE Timing Goes Negative

07:27 Defense Plan and Deadline