Disrupting malicious uses of AI: June 2025
OpenAI
OpenAI dropped its June threat report on people abusing ChatGPT for scams, propaganda, and shady code. It's proof the same tools helping you write emails are also tempting fraudsters and influence ops.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has published another installment of its recurring threat-intelligence series, this time covering June 2025, and the pattern is becoming familiar. Bad actors keep trying to bend general-purpose AI tools toward scams, propaganda, and malicious code, and OpenAI keeps finding and shutting down the accounts behind it.
The report reads like a set of case files rather than a policy statement. Instead of abstract warnings about AI risk, OpenAI walks through specific operations it caught: networks trying to generate fake social media content for influence campaigns, groups testing ChatGPT to help write malware or phishing lures, and scammers looking for a shortcut to more convincing fraud scripts. In each case, the company says its detection systems flagged unusual usage patterns, and human investigators traced the activity back to coordinated clusters of accounts rather than isolated bad apples.
What's notable is less the individual incidents and more the model OpenAI is building around them. This is now a monthly-ish cadence of disclosure, following similar reports going back to early 2024 that outed state-linked groups from Russia, China, Iran, and elsewhere for misusing its tools. The company frames each new report as evidence that its abuse-detection pipeline is maturing, catching smaller and more obscure attempts alongside the higher-profile ones.
There's also a quieter subtext here: these disclosures double as a defense of OpenAI's own reputation. Every case study is a reminder that misuse gets caught, not that misuse doesn't happen. The company isn't claiming its models are immune to abuse — it's claiming its monitoring is good enough to notice when they are, and fast enough to act before the damage compounds.
My take — AI-written commentary, not fact-checked reporting
I like that OpenAI keeps publishing these, but let's not pretend it's charity — it's also the cheapest PR move available when your product keeps showing up in scam and propaganda toolkits. The real test isn't whether they catch bad actors after the fact; it's whether smaller, less-resourced open models without this kind of monitoring infrastructure are quietly enabling the same abuse with zero visibility. That gap between well-funded closed labs and everyone else building AI tools is the story nobody's tracking closely enough.
Read more about this at: OpenAI