OpenAI Shut Down an Israel-Linked AI Influence Operation Called "Zero Zeno"!
Hi, it's Shiichan! Today I've got a story about AI misuse getting caught and shut down, and it's a bit of a wild one.
OpenAI News
openai.com
What was announced?
OpenAI published a report called "Operation Zero Zeno" on its News page. It describes a network of accounts run by STOIC, a political campaign management firm based in Israel, that OpenAI discovered and banned. The name honors Zeno, the founder of Stoic philosophy, and doubles as a nod to just how little engagement the campaign actually got.
Why it matters
OpenAI regularly investigates whether its models are being abused for influence operations, and it publishes what it finds. Zero Zeno used AI to generate articles and social media comments in bulk and post them across multiple platforms.
It initially targeted audiences in Canada, the U.S., and Israel with English and Hebrew content, expanded to English-language content aimed at India starting in May, and appeared to be preparing a campaign for Ghana too. It's a reminder that AI-generated propaganda can span far more countries and topics than you'd expect from a single operation.
What changes
By publishing this report, OpenAI has already disrupted the accounts and sites STOIC was using. Just as important, the report lays out exactly how the operation worked: the fake personas, the self-replying accounts, the recycled profile photos, the account-cycling after bans. That gives other platforms and security teams a concrete pattern to watch for. It's a small but meaningful step toward pushing back on AI-powered fake engagement.
Dive Deep
Here's how the operation worked:
- Generated articles and social comments in bulk with AI models and posted them across platforms
- Built fictional personas with varied age, gender, and location details
- Used other accounts to reply to its own posts, faking the appearance of real engagement, sometimes reusing the same GAN-generated profile photo across accounts
- Created new accounts to keep posting whenever platforms banned the old ones
Where it posted:
- X (formerly Twitter)
- YouTube
- Four websites posing as activist groups (nonagenda.com, the-good-samaritan.com, uc4canada.com, ufnews.io)
Themes it pushed:
- Criticism of the UN relief agency UNRWA
- Criticism of "radical Islamists" in Canada
- Criticism of pro-Palestinian activism at U.S. universities
- Criticism of Qatar's investments in the U.S.
- Content defending Israel's Histadrut trade unions
- Election-related content criticizing India's BJP
Despite all that effort, the actual reach was tiny. OpenAI rated the operation Category 2 on its Breakout Scale (low real-world spread), noting:
"Zero Zeno's activity appears to have attracted little if any engagement, other than from its own inauthentic accounts."
Some of its YouTube videos reportedly got zero views.
Wrap-up
- STOIC, an Israeli political campaign firm, used AI to mass-produce influence-operation content
- Targets included Canada, the U.S., Israel, and India (English/Hebrew), with prep work underway for Ghana
- Content spread across Instagram, Facebook, X, YouTube, and four fake activist websites
- The tactics were elaborate (fake personas, self-replies, recycled photos), but real-world reach was close to zero
If you work in trust & safety or security research, or you're just curious how AI-powered influence operations actually play out (and often flop), this report is worth a read!
