OpenAI has banned a cluster of Russian-linked ChatGPT accounts used to run a covert influence operation at scale — one that relied on AI to build a fake think tank, manufacture articles, and flood social media with fabricated content. The operation, tied to a fictional organisation called the International Burke Institute (IBI), ran for months before OpenAI detected and disrupted it.

This is not a theoretical risk. It happened. And it tells businesses a great deal about where AI-powered deception is heading.

What the Operation Did

The actors behind this campaign set up what appeared to be the International Burke Institute, a self-described expert community supposedly based in Israel. The IBI had a website, published articles attributed to named experts, and ran a so-called “sovereignty index” ranking countries on governance — predictably praising Russia while targeting France, Germany, and the United States.

ChatGPT was used to:

• Generate English-language social media posts distributed across Substack, Telegram, X, Facebook, and LinkedIn
• Produce or repackage articles under the IBI banner
• Create content that masked its Russian origin, hiding linguistic markers that might have revealed the operators’ location

The operators used VPNs to access OpenAI’s platform, since ChatGPT is not available in Russia. Once inside, they reportedly instructed the model to avoid any phrasing that sounded Russian in origin.

When OpenAI reviewed a sample of 36 articles linked to IBI experts — published between September 2025 and May 2026 — they found 34 had been lifted or adapted from other sources. Some were misattributed entirely, including content falsely credited to well-known academics like Francis Fukuyama and Noam Chomsky.

Why This Matters for Businesses

AI makes disinformation cheaper, faster, and harder to detect. What previously required a room full of writers and editors can now be produced by a handful of operators with API access and a clear objective.

A few things stand out here that matter directly to organisations using AI tools:

AI content has a trust problem it didn’t create. The more widely AI-generated content spreads, the harder it becomes for buyers, partners, and journalists to distinguish authentic thought leadership from manufactured noise. Businesses that publish AI-assisted content need to think harder about provenance, human review, and disclosure.

Enterprise platforms are being weaponised at the edges. This campaign did not break OpenAI — it exploited standard API access, VPN obfuscation, and the absence of human review. The attack surface is not the model; it is the workflow around the model. Enterprises deploying AI tools at scale face the same exposure.

Detection is getting better, but slowly. OpenAI identified and disrupted this campaign, but not in real time. The IBI ran for months. Organisations relying on AI content detection to protect their brand or verify third-party claims should assume detection lags behind creation by a meaningful margin.

What OpenAI’s Response Shows

Disrupting an operation like this requires a combination of account monitoring, content fingerprinting, and network analysis — looking at patterns across accounts rather than evaluating any single piece of content. OpenAI’s ability to detect and shut this down is a capability step forward.

But it also raises a question: how many smaller or less organised operations are running under the same threshold? The IBI was sophisticated enough to build a persistent presence across multiple platforms. Less sophisticated variants might never get flagged.

What This Means for Business Leaders

If your organisation is publishing content, evaluating vendors, or relying on expert citations, AI-generated misinformation is a growing due diligence problem. A few practical considerations:

• Verify citations and sources independently. If an article or report cites a think tank you have not heard of, look past the name. Check whether the institution has real funding, real staff, and a credible publication history.
• Audit your own AI content workflows. If AI is drafting content for your business, a human review step is not optional. This is not about the model’s quality — it is about accountability for what carries your name.
• Train your team to be sceptical of online authority. A LinkedIn profile with a thousand followers and a plausible employer is not the same as a verified expert. That has always been true; AI makes it more relevant.

For businesses building out AI capabilities, the lesson is not that AI is dangerous. It is that AI deployed without governance is exploitable. The same tools that help your team move faster can help bad actors manufacture credibility at scale.

OpenAI’s intervention here is a signal that AI providers are taking platform misuse seriously. It does not mean the problem is solved.

---

Enterprise DNA works with organisations to build AI systems that are reliable, governed, and aligned with business outcomes rather than short-term productivity gains. If you are thinking about how to deploy AI responsibly in your organisation, start a conversation with our team.