AI for Work

Operation "No Bell": A coordinated chorus of criticism against the US and its allies

OpenAI disrupted a covert influence operation, nicknamed "No Bell," reportedly originating in Russia, which used generative AI to mass-produce criticism of the United States and its allies targeting audiences across Africa. The case illustrates how AI lowers the cost of propaganda, why African nations are increasingly targeted by influence campaigns, and highlights the growing responsibility of AI platforms to detect and ban coordinated deceptive activity.

If you follow how artificial intelligence is being misused for influence campaigns, a new case out of OpenAI deserves your attention. The company recently shut down a set of accounts tied to a covert operation—apparently originating in Russia and not previously reported publicly—which researchers have nicknamed "No Bell." The network relied on AI tools to generate criticism of the United States and its partner nations, aiming that content at readers throughout the African continent. This article explains what happened, why covert campaigns like this matter, and what lessons everyday users and content platforms can take away from it.

What OpenAI Discovered

OpenAI's trust and safety teams regularly review how its models are being used, looking for patterns that suggest coordinated, deceptive activity. In this instance, investigators noticed clusters of accounts behaving in ways consistent with an influence operation rather than genuine individual use.

The network, given the internal name "No Bell," was reportedly using generative AI to draft and polish written material attacking US foreign policy and the interests of American allies. The intended audiences were located across multiple African countries, suggesting a deliberate regional targeting strategy rather than random distribution.

What made this case notable was the likely origin. OpenAI assessed that the operation probably came from Russia, although the company stopped short of definitive attribution—covert campaigns are designed precisely to hide who is behind them. Because the activity had not been reported before, OpenAI's disclosure offered the first public window into this particular effort.

How AI Fits Into Influence Campaigns

Influence operations are not new. Governments and political groups have long used fake social media personas, staged news sites, and paid commentators to sway opinion abroad. What has changed is the tooling.

Generative AI lowers several barriers at once. First, it solves the "volume" problem: producing dozens of plausible articles, social posts, or comment-thread replies takes minutes instead of hours. Second, it reduces the "quality" problem, because text comes out fluent and reasonably natural-sounding even when the operators are not native speakers of the target language. Third, it allows quick adaptation—a single campaign can produce variations of the same message tailored to different countries or platforms.

In the No Bell case, AI appears to have been used as a writing assistant for propaganda content, not to create deepfake videos or voice clones. That distinction matters, because it shows that even plain text generation can be enough to make a covert campaign cheaper and faster to run.

Why Africa Is a Target

Foreign influence efforts have increasingly focused on African nations for strategic reasons. Several governments on the continent are weighing relationships between Western powers and other major players, and information campaigns attempt to nudge public sentiment in one direction or another.

For operators behind campaigns like No Bell, Africa represents both an audience that other influence efforts have historically underinvested in and a region where local media ecosystems can be more vulnerable to coordinated messaging, since fact-checking infrastructure varies widely between countries. Targeting criticism of the US and its allies fits a familiar pattern: rather than promoting one side's virtues, these campaigns often concentrate on undermining trust in rival powers.

None of this means African audiences are uniquely gullible—researchers repeatedly find that most people encounter such content with skepticism. The concern is scale: when networks flood platforms with coordinated messages, they can create a false impression of widespread opinion.

What This Says About AI Platform Responsibility

OpenAI's decision to ban the accounts highlights the role AI providers now play in the information ecosystem. Since the company can observe how its models are actually being used, it is often among the first to detect coordinated abuse—sometimes before social media platforms or independent researchers see the downstream content.

The company has a stated policy prohibiting the use of its tools for influence operations and deceptive political activity. Enforcing that policy means tracing patterns across accounts: repeated prompts on similar themes, shared writing styles, links to the same distribution channels, or timing patterns that suggest a small team running many identities.

It is worth being honest about the limits. Banning accounts disrupts a campaign, but operators can simply move to another provider or rely on open-source models. Detection and enforcement raise the cost of abuse without eliminating it. That is why researchers generally argue for a layered approach, involving AI companies, social platforms, fact-checkers, and newsrooms working in parallel.

What Readers Can Do

For ordinary readers, cases like No Bell are a reminder to treat politically charged online content with healthy caution:

  • Check the source. Covert campaigns often hide behind newly created accounts or websites with no track record.
  • Look for corroboration. If a striking claim about US or allied policy appears only on obscure outlets, that is a signal to verify elsewhere.
  • Notice uniformity. If many different accounts push nearly identical talking points at the same time, coordination is likely.
  • Be skeptical of outrage. Content engineered to provoke anger travels fastest and is a favorite tool of influence operations.

None of this requires technical expertise—just a habit of pausing before sharing.

The Bigger Picture

The No Bell takedown fits into a broader trend: state-linked and politically motivated groups are adopting generative AI faster than many predicted, while the defenses remain largely reactive. Expect more disclosures of this kind from OpenAI and its peers, and expect campaigns to keep evolving in response.

The reassuring part is that AI-generated propaganda still tends to be detectable—it often repeats itself, lacks local nuance, and connects back to the same distribution networks. Vigilance from platforms, researchers, and readers together is what keeps these campaigns from achieving their goals.

Comments (0)

  1. No comments yet. Be the first to share what worked for you.

Leave a comment

Comments are reviewed before they appear. Your email address is not published.