←
Ben Nimmo8,332 posts
Ben Nimmo avatar

Ben Nimmo

@benimmo

Writer, linguist, diver. Principal Investigator, Intelligence & Investigations @OpenAI. RT ≠ endorsement.

◷ Joined July 2009▧ 2,953 Photos and videos
Ben Nimmo avatar
Ben Nimmo@benimmo·16 Mar 2023

NEW: Meet the Online Operations Kill Chain, a new cross-threat framework to analyse online ops: from influence ops to cyber espionage, spam and more. h/t my co-author @killchain and thanks to @IOpartnership, @Meta + industry teams for feedback/expertise!

0:35
Ben Nimmo avatar
Ben Nimmo@benimmo·21 Feb 2025

3) Two scam centres in Cambodia, one focusing on romance scams, the other on task scams. We found one through proactive investigation. H/t @meta for alerting us to the other.

Post attachment
Ben Nimmo avatar
Ben Nimmo@benimmo·21 Feb 2025

2) Also likely China-origin covert IO we called “Sponsored Discontent”. It generated tweets criticizing dissident Cai Xia, but also translated articles from Chinese to Spanish and landed them in news outlets in LATAM and Spain. One of the articles was listed as “sponsored”.

Post attachmentPost attachment
Ben Nimmo avatar
Ben Nimmo@benimmo·21 Feb 2025

There’s a lot to dig into here. Highlights: 1) Likely China-origin surveillance operation we dubbed “Peer Review”, using our model to modify and debug code for a social media monitoring tool apparently run off an open-source AI model.

Ben Nimmo avatar
Ben Nimmo@benimmo·21 Feb 2025

These disruptions included operations from around the world, including likely activity from: China 🇨🇳 Iran 🇮🇷 UAE and Ghana 🇦🇪 🇬🇭 Cambodia 🇰🇭 North Korea 🇰🇵

Ben Nimmo avatar
Ben Nimmo@benimmo·21 Nov 2024

We're dedicated to identifying, preventing, and disrupting attempts to abuse our models, and welcome Meta’s sharing of threat information about the scam centers they have disrupted.

Ben Nimmo avatar
Ben Nimmo@benimmo·21 Nov 2024

This included producing comments to share on social media accounts, which appeared deceptive and consistent with scam activity.

Ben Nimmo avatar
Ben Nimmo@benimmo·21 Nov 2024

As part of our proactive investigative work, we recently detected and banned a cluster of ChatGPT accounts that appeared to be originating in Cambodia and engaging in translating and generating comments.

Ben Nimmo avatar
Ben Nimmo@benimmo·9 Oct 2024

Final point: none of the operations we identified so far looked like it achieved viral breakout or audience engagement because of its use of AI. Social media are a tough environment. But we’ll keep hunting, and keep reporting what we find.

Ben Nimmo avatar
Ben Nimmo@benimmo·9 Oct 2024

Some of the operations we disrupted and reported on here were first exposed by other investigative teams. Some, as far as we know, were previously unreported. Some of them, we discovered *because* we use AI in our investigations.

Ben Nimmo avatar
Ben Nimmo@benimmo·9 Oct 2024

That means AI investigators have a different view of threat actors - not necessarily earlier or more granular, but different. It means we can sometimes add to the defender community’s knowledge of existing threats, or identify new ones.

Ben Nimmo avatar
Ben Nimmo@benimmo·9 Oct 2024

AI companies, and thus AI investigators sit in a unique niche in the information space - midway between upstream providers of things like emails, and downstream distribution platforms like social media.

Ben Nimmo avatar
Ben Nimmo@benimmo·9 Oct 2024

Plenty of individual case studies for aficIOnados, but there’s an interesting and overarching point here that came out of our research, as we analysed 20+ disruptions this year: where AI companies sit in the overall investigative landscape.