OpenAI Agents Hit UN Trade Site 16,000 Times in Three Months, Researcher Says

OpenAI's autonomous agents repeatedly hammered a United Nations statistics website this spring, according to security researcher Rowan Howard-Jones, who says the tools scanned the UN Conference on Trade and Development's data portal more than 16,000 times between April and June. The activity was not a single burst but a sustained campaign of requests against UNCTAD's public statistics infrastructure, the kind of pattern that site operators typically associate with hostile traffic rather than legitimate research.

Howard-Jones characterized the incident as an attempt to 'bruteforce' the site, though he was careful to note it does not match the severity of the Hugging Face breach or the recent wave of attacks on US government domains. The distinction matters: UNCTAD's statistics pages are public, so the agents were not breaking into anything. What concerns researchers is the volume and persistence of automated requests coming from a major AI vendor's tooling, which can look indistinguishable from a denial-of-service attempt to the engineers watching server logs.

The episode fits a broader pattern that has emerged over the past year as AI companies ship agents capable of browsing, clicking and filling out forms on their own. Those systems are designed to complete tasks efficiently, and when a page is slow or returns an error, many are built to retry, sometimes aggressively. Site operators have reported similar traffic spikes from other crawlers and agent frameworks, though few have been documented with the specificity Howard-Jones attached to the UNCTAD case.

For OpenAI, the report lands at an awkward moment. The company has spent much of 2025 promoting agents as productivity tools for enterprises, and reliability under real-world conditions is central to that pitch. A sustained run of requests against a UN site gives critics a concrete example to point to when they argue that agent guardrails, rate limits and disclosure practices lag behind the capabilities being shipped.

What remains unclear is whether the traffic originated from OpenAI's own infrastructure or from third-party developers running the company's models, and whether UNCTAD noticed or complained at the time. Howard-Jones has not published a full technical writeup, and OpenAI has not responded publicly to the claim. The next signal to watch is whether the company adjusts rate limiting or retry behavior for its browsing agents, and whether other site operators come forward with similar logs.