OpenAI Shuts Down Iranian Election Interference Scheme - The AI Podcast

Transcript

A (0:00)

OpenAI has said that they have just shut down an Iranian election meddling kind of campaign. This is so fascinating. There's a bunch of, you know, there's a bunch to this story that goes from Microsoft that did a lot of the investigative work to how they kind of stopped it, to what the results were, to who I ran or this group was campaigning for. There's so many interesting I want to break down for you today. Before we do that, I want to say I have just launched my very first podcasting course. Now, if you have ever wanted to start a podcast or if you have a podcast and you want to learn ways to make it more professional and to scale and grow it, I have a podcast course that breaks down the exact steps on how I got over 4 million downloads on my podcast and how I was able to essentially grow the audience, how I was able to do the content. I have an entire podcast studio now and a studio manager and someone that helps him with this content. I was able to essentially build this machine. Whether you want to do it the most professional way or you want to get started, or do it in an amateur way. I have a podcast course that covers all of that and especially how to grow your podcast. So there's a link in the description podcast studio.com courses where you can go learn a little bit more about this course. I would love to have you join. There's a discount for this week only, so go and check out the course. It's linked below. Now let's get to the podcast. The thing that I want to say that is really interesting is that. So there was A recently a TechCrunch article that kind of, you know, covered what's going on specifically with this entire story. But OpenAI has also published a blog post that's really interesting on this. Now, the thing that is interesting is essentially OpenAI, they banned a bunch of different ChatGPT accounts. This is kind of what they have said they're going to be doing is like, if they notice this is happening, they're going to go and ban all of the accounts related to it. Now, a lot of people have said, like, how effective is this? Are they going to be able to get around this? I actually think that there is an effect, there is some level of effective to this. So essentially, if they detect people are doing this when it says ban accounts, sure. People are still going to be able to go to ChatGPT and say, like, write, you know, write a social media post from a user about how bad Donald Trump is. Write a post from Social media user about how bad Kamala Harris is. Right. Like, anyone could still do that. And so technically, like if they're trying to like appear like there's a whole bunch of people with a specific narrative, they could use ChatGPT, maybe free accounts or not logged in accounts or random accounts to kind of do this sort of stuff. Once they start doing it en masse, that's typically when they're actually using an API. So they have to have a specific business set up for it. They have to have a credit card on file. And as soon as that whole scenario happens, it's a lot, it's a lot more difficult to set the accounts up. So when they do ban an account for, with that has like an API, there's, there's a lot more they can do to actually stop those actors. Now will these people just like, oh, they banned us, we're giving up? No, they'll go use anthropic, they'll go try to make new accounts. They'll go use like all the other AI models. Like, this isn't over. But I think it's interesting because it shows what's going on and hopefully this slows them down a bit. To me, I'm actually more interested in the techniques they were using, who they were campaigning for, what they were saying. I think there's a lot that we can learn because they actually have what the websites were, they have screenshots, they have everything on this. So this is very interesting. If you go over to ChatGPT's blog post they just released on this, it's called Disrupting a Covert Iranian Influence Operation. I think this is kind of like a feel good story for them to publish in a way, because this covert operation or whatever was campaigning both for and against liberals and conservatives. So they, it had accounts that were pro conservative and it had accounts that were pro liberal. And I think that this is, honestly, I think this is a really smart strategy. It's not to just align with one specific political ideology, but if you're a foreign country, there's probably some things that you like from both parties, different policies that you might like from both parties. And if nothing else, their goal is to kind of stoke division in America, you know, weaken your enemy. That, that's one concept. The other concept is they, they can essentially build a following on the right and a following on the left politically. And then they can have that same account that like traditionally always says pro conservative things. All of a sudden, maybe say something less conservative, but that that country specifically likes and same thing with liberals always says liberal things and then something less liberal that benefits a specific country. Now you can imagine, as with Iran and the demographic there, a lot of the content around this actually related to Israel and, and Palestine and Gaza and that whole situation. So I did find that very interesting. Now I just wanted to say this is something that ChatGPT has been talking about for a while. Back In February of 2014, or, sorry, February this year, February 14th, they had a, a blog that was called Disrupting Malicious Use of AI by state Affiliated Threat Actors. And they talked about the fact that they were looking into this, this was something they were taking seriously and how they were kind of going to be, you know, what their multipronged approach to AI safety was going to be in this regard. So this is something they've been like, you know, they have a blog post out since February and now they're, they're, you know, they have a specific operation that they're reporting on. Now the thing I want to say about this whole operation before we get into it and what they did and what they found was that this was something that they actually. So, you know, they, they identified the cluster of ChatGPT accounts. They were making all of this content. It apparently was a, something called Storm 2035. But this was actually something that Microsoft discovered a week ago and reported on and had like a whole report. And then OpenAI, I think, went and like internally looked at, you know, if any of the content that was being generated around that was from their platform and they found it was. So they, they took steps. We'll talk about their steps. But I want to talk about this storm 2035 really quick. What's really interesting here. Microsoft published a whole thing on August 9th on this, but they said, quote, foreign malign influence concerning the 2024 US election started off slowly, but has steadily picked up the pace over the last six months due initially to Russian operations, but more recently from Iranian activity. This third election report from the Microsoft Threat Analysis center provides an update on, you know, what we've observed from Russia, Iran and China since our second report in April of 2024. Okay, so they've done a number of reports. They talked about, you know, how Iran was kind of entering the fold with cyber influence operations starting in June. I think this is interesting. They said past Iranian election influence efforts and they've talked about in Israel, apparently they've kind of done some things. In Turkey, they've done some things, um, and then also in the United States last year, apparently they had some interesting things where I think in, in the US specifically last year they did SMS and email, traditional media impersonation and some doxing and some, some, some content application. So that was interesting. It seems like they're getting a little bit more sophisticated this year. So what I did think was interesting though is kind of looking over what they were posting about. You know, there's an Iran run covert news site targeting US voter groups which has four websites that it says are masquerading as news outlets. Engaging US voter groups on opposing ends of the political spectrum with polarizing messages on issues such as US Presidential candidates, LGBTQ rights and the Israel Hamas conflict. The Microsoft said this group is part of a broader campaign that has been operating since at least 2020 and includes over a dozen covert news sites targeting French, Spanish, Arabic, English speaking audiences with social and political content. This is interesting. My one comment on this and I know that this is like a real thing that foreign actors are trying to influence in elections and like content, whatever, but I also kind of wonder how, I wonder like what the big difference between this is versus like state run media. So in China we obviously have like I think all the news is state run, right? Specifically like the South China Morning Post. It's like a Chinese based news site but like it definitely everything is kind of pro China that would come out of that. I feel like me being from Canada, I feel like it's kind of similar with the BBC or the oh cbc. That's like our big news site is the CBC Canadian Broadcasting and it, I think it receives like a billion dollars a year in like taxpayer funding grants from the government and it is currently, it has received a lot of that funding from the current administration and it's very positive towards the current administration as you can imagine. I think it'd be bankrupt anyways, like not to get into too much like politically but I just wonder like how some of these state run. I think in Russia there's like RT is like the big Russia state run kind of or influenced one. I wonder like what the big difference between like RT and CBC and you know, South China Morning Post, like what the big difference between those. And like these are like these are obviously like covert random news groups because if you just had like the Iranian times people would be like oh, I don't know if I'm going to trust that's an unbiased source. Whereas if it's like the Canadian Broadcasting Group it's like okay, it definitely is going to be biased towards Canadian things or South China Morning Post definitely going to be biased towards Chinese, you know, interests or whatever. So I guess maybe that's. Maybe that's the thing. Or the Russia times obviously is going to be biased towards Russian things. So I guess maybe that's part of it. But that's on the news front now they're doing something completely different, which obviously is quite cynical on social media. The social media front, which is they have a ton of different accounts that they've created on X and I think Instagram, maybe other social media sites where they have these fake users essentially commenting to make it seem like one narrative is more or less popular and essentially putting their opinion in. Now, this is something that is quite easy to do, right? You essentially have almost like agents, but they're. And I think this is an important issue to think of as we get more and more closer to like real agents running completely autonomous. But you essentially have these accounts and anytime there's a viral post about something, they could have two different people commenting on it from two different political perspectives to try to get users on those accounts from two different political sides. And then they use those accounts to push whatever their own agendas are, whatever their own messages are. And that doesn't necessarily mean it has to push. I think, like a lot of times, because of, like the last election cycle, there's, you know, a lot of people saying, I think, I mean, I don't know, maybe this is. Maybe the people. This is controversial. People disagree with me. I feel like from the news coverage I've seen, Republicans in America are more likely to accuse liberals of siding with China, and liberals in America are more likely to accuse Republicans of siding with Russia. So I don't know if that's 100% accurate. I'm sure everyone's got. I'm sure, like that I probably. I hopefully offended everyone equally on that. But I mean, this is just like the narrative I see as a Canadian more so I don't know, whatever. If you're mad about it. Sorry. This is just like what I feel like I see more. But regardless, it's kind of interesting because all of those different groups have their own, you know, ideas and stuff. And so they're essentially, if they can grow a following in one specific political spectrum, it's not just about which political party wins, but it's about the policies. So there. I think the goal for them is to push the policy. They don't really care what political party is in power. Right. And I'm sure all Republicans say, oh, no, all of our adversaries want liberals. And I'm sure all Liberals say oh no. You know, all Democrats say oh no. All of our foreign adversaries want the Republicans in control. I'm sure that's. But I mean, like, let's just idea what, let's try to be intellectually honest and say that probably foreign actors don't care who's in charge as long as they're doing a bad job and supporting their regime or their, you know, policies that they think benefit them. Right. Like I, I think that that's, I'm trying to be as unbiased. That's probably pretty intellectually honest. So let's try to just be unbiased there. So it doesn't, it probably doesn't matter as much. And this is what we're essentially seeing from these, these groups is that they, they try to stoke division on both sides. China is, is, you know, stoking division on both sides. Iran is stoking division. Russia seems to be stoking division. It doesn't necessarily matter. So as far as this Iranian threat goes, OpenAI has said in kind of relation to this that they said, quote, our investigation revealed that this operation used ChatGPT for two purpose, generating long form articles and shorter social media comments. The first workstream produced articles on US politics and global events published on five websites that posed as both progressive and conservative news outlets. The second work stream created short comments in English and Spanish which were posted on social media. I think this is interesting. They're doing, you know, they know it's, it's just like an English thing. They're like, well, you know, there's a huge demograph of Spanish speaking populations and other populations. They're doing every language. Which I think is also very interesting. They said, we identified a dozen accounts on X and one on Instagram involved in this operation. Someone, some of the X accounts posed as progressives and others as conservatives. They generated some of these comments by asking our models to write comments posted by other social media users or to rewrite comments. So essentially they would go to Twitter or they would go to Instagram, find a comment and be like, hey, rewrite this comment. So it's like a pro Trump comment or pro Kamala comment and they would just have a rewrite the comment or negative to either one. Which is interesting because, you know, it's not like they have to come up with an elaborate prompt of like how to write the perfect political comment. They just go and write, rewrite other people's comments. Someone else is already doing all the work for them. They just say it in a different way. And People are, you know, gonna like it. And I also think that's a good strategy because they could go to posts and see, like, what the top five comments are with the most likes. And they're like, okay, these comments obviously get a lot of engagement. Rewrite these ones, write them as our own. Try to get followers, like, similar followers to our account. So they have two screenshots that they. They showed to kind of show that I think this is not a partisan. This should be a bipartisan issue. Everyone should be concerned about this. The first one was an article that, you know, this, you know, network created called why Kamala Harris picked Tim Waltz as her Running mate. A calculated choice for unity. And the second one was X censors Trump's tweet, a hidden war on free speech. And so, you know, one of them obviously is pro Kamala, and the other one is, you know, pro Trump, and, you know, why he got mistreated or whatever. So these appeal to totally different audiences, which I think is so interesting. They also showed two different social media posts. One of them said, Kamala Harris's policies are criticized for potentially leading to increased immigration costs and providing citizenship. Critics argue she might attribute these changes to climate change. And it's like rolling its eyes. And then it says, hashtag dump comma. Okay? So this is like an anti comma one. Then they had another one that was generated that says, for 248 years, Americans have been warned about someone like Donald Trump trying to crown himself king. I'm ready to fight like hell to protect our democracy. Hashtag dump Trump. Okay, that was the other one. So I think this is interesting, right? We're seeing completely different political ideologies being promoted. And then they also have, like, a bunch of posts, Spanish and English, that are just, like, about beauty content and kind of random content. So I think it's trying to just, like, get more followers in this regard. But regardless, I think this is interesting. They. They also outline a number of other areas that this group has tried to influence. So it says, quote, the operation generated content about several topics, mainly the conflict in Gaza, Israel's presence at the Olympic Games and the US Presidential elections, and to a lesser extent, about politics in Venezuela, the rights of Latino communities in the US Both in Spanish and English and Scottish independence. They intersperse their political content with comments about fashion and beauty, possibly to feel appeal more authentic or in attempt to build a following. So I think this is interesting. What they've essentially said in this blog post, though, is that regardless of all of the. The comments and stuff. None of the social media accounts that they were able to track down and link to this. I had a ton of engagement. Now I think that that doesn't necessarily mean that they found all of the accounts or maybe they just found them early on. I don't know. It also doesn't mean that like they couldn't have been using anthropic or other accounts couldn't have been discovered that I don't know that do similar things. So I don't think it just means like we're totally out of the woods yet when they're just like, look, we look, we check down the social media accounts and they don't have a ton of engagement. I don't know if that really means we're like totally clear on this. I do think this is interesting. I mean, from their perspective, sure, this is a good strategy. You go kind of stoke division in the United States who might be your foreign adversary, or you try to build followings on both sides politically and then try to influence public policy or opinion to certain, you know, in regards to certain conflicts. So I think this is an interesting strategy. We're starting to see roll out. This is going to be something that becomes more and more in like this is going to be a much bigger deal in the future as I think we have completely autonomous agents that can run some of these tasks. Not very complicated stuff, right? You write scripts that can go to social media, grab comments, repost them and then kind of add your own ideologies and stuff in there, tweet at people, like do all sorts of things. There's a lot going on on Twitter right now where people are accusing different accounts of being AI agents or AI run. Maybe they're early versions of GPT5 that got leaked, whatever. There's a bunch of, you know, controversy and stuff. But I think this is definitely a topic to be aware of as this is going to be a bigger issue in the future. So if you enjoyed the episode, if you learned anything, it would mean the world to me if you could write a review. And if you are interested in my podcasting course where I cover exactly how to go from zero to a hundred podcasting, or if you already have a podcast, I want to take it to the next level. I talk about, you know, from the most beginner setup I had, um, to kind of my amateur podcasting setup to when I went and had a full on studio with a production crew to help me produce my, my content. So everything I've done at all phases of this podcasting journey and all of the marketing tips and strategies I use to organically grow to over 4 million downloads. It's covered. It's all it's covered in the course. If you're interested in starting a podcast or you've ever wanted to, it's on. It's it's on sale this week only. You can go check out the link in the description podcast studio.com course. Hope that you all have an amazing rest of your day.

The AI Podcast: Episode Summary - "OpenAI Shuts Down Iranian Election Interference Scheme"

Release Date: November 14, 2024
Host: The AI Podcast
Episode Title: OpenAI Shuts Down Iranian Election Interference Scheme

1. Introduction to the Election Interference Scheme

The episode delves into a significant development where OpenAI has successfully dismantled an Iranian campaign aimed at meddling in the 2024 U.S. elections. The host emphasizes the complexity of the operation, highlighting the collaborative efforts between OpenAI and Microsoft in uncovering and mitigating the threat.

Host [02:15]: "OpenAI has just shut down an Iranian election meddling campaign. This story involves extensive investigative work by Microsoft and showcases how they collectively addressed the threat."

2. OpenAI’s Response and Action Taken

OpenAI responded to the interference by banning multiple ChatGPT accounts associated with the Iranian scheme. The company's proactive measures are part of their broader strategy to combat malicious use of AI.

Host [10:05]: "OpenAI has announced that they will ban all related ChatGPT accounts once they detect such activities, aiming to curb the spread of coordinated misinformation."

Despite skepticism about the effectiveness of account bans, the host believes that OpenAI's actions have a measurable impact, particularly against sophisticated actors using APIs for mass content generation.

3. Microsoft’s Role in Investigative Efforts

Microsoft played a pivotal role in identifying and reporting the malicious activities. Their threat analysis provided crucial insights into the operations of foreign adversaries seeking to influence U.S. elections.

Host [15:40]: "Microsoft published a comprehensive report on August 9th, detailing the evolution of foreign influence operations, initially driven by Russia and recently amplified by Iran."

4. Overview of Storm 2035

The interference campaign, dubbed "Storm 2035," represents an escalation in Iran's cyber influence operations targeting the U.S. The operation employs sophisticated techniques to disseminate polarizing content across various digital platforms.

Host [22:30]: "Storm 2035 is part of a broader campaign that's been active since 2020, targeting diverse audiences with content in multiple languages to amplify division within the U.S."

5. Techniques Employed in the Campaign

The Iranian scheme utilized ChatGPT to generate both long-form articles and short social media comments. These were strategically posted on fake news websites and social media platforms to sway public opinion on critical issues.

Content Generation: Creating articles on U.S. politics and global events, masquerading as either progressive or conservative news outlets.
Social Media Manipulation: Developing comments in English and Spanish to appear authentic and engage with different political spectrums.

Host [30:50]: "The operation used ChatGPT to produce articles and comments that mimic genuine user interactions, effectively blurring the lines between authentic discourse and orchestrated propaganda."

6. Impact on Public Discourse

The campaign aimed to stoke political division by presenting biased narratives from both ends of the political spectrum. By doing so, the adversaries intended to weaken societal cohesion and influence policy-making irrespective of which political party was in power.

Host [35:20]: "By fostering division on both sides, the goal isn't necessarily to sway the election outcome but to create a more fragmented and less resilient political landscape."

7. Comparison with State-Run Media

The host draws parallels between the covert Iranian operations and traditional state-run media outlets like RT (Russia Today) or the South China Morning Post. However, he notes that while state-run media typically exhibit overt biases aligned with their respective governments, the covert operations employ more nuanced and deceptive tactics to appear unbiased.

Host [40:15]: "Unlike clear-cut state-run media that openly support their government's agenda, these covert groups blend in by mimicking neutral news sources, making their influence harder to detect."

8. Challenges in Mitigating AI-Driven Interference

Despite OpenAI's efforts, the host acknowledges the ongoing challenges in completely eradicating such interference. He points out that adversaries may shift to other AI models or create new accounts to continue their operations.

Host [45:00]: "Even with account bans, these actors are likely to migrate to other platforms or AI models like Anthropic, ensuring that the battle against misinformation remains continuous."

9. Future Implications and AI Safety

The episode underscores the growing importance of AI safety and the need for robust strategies to combat automated misinformation campaigns. The host anticipates that as AI technology advances, so will the sophistication of such interference tactics.

Host [50:30]: "As we approach an era with autonomous AI agents, the potential for more intricate and widespread interference operations increases, necessitating forward-thinking AI safety measures."

10. Conclusion and Reflections

In wrapping up, the host reflects on the significance of OpenAI's actions in the broader context of AI governance and the fight against malicious use of technology. He emphasizes the need for continued vigilance and collaboration among tech companies to safeguard democratic processes.

Host [55:45]: "OpenAI's proactive stance is commendable, but it's just one piece of the puzzle. A collective effort is essential to ensure that AI serves as a force for good rather than a tool for manipulation."

Key Takeaways:

Collaboration is Crucial: The joint efforts of OpenAI and Microsoft highlight the importance of partnerships in combating AI-driven threats.
Sophistication of Modern Interference: The use of advanced AI models like ChatGPT allows for more subtle and effective manipulation of public opinion.
Ongoing Battle Against Misinformation: Despite measures taken, the fight against malicious use of AI remains relentless as adversaries adapt and evolve their tactics.
Future of AI Safety: As AI technologies become more autonomous, establishing comprehensive safety protocols and strategies is imperative to prevent their misuse.

This episode provides a comprehensive overview of the recent Iranian election interference scheme, offering insights into the methods used, the challenges faced in combating such threats, and the broader implications for AI safety and governance.