Disrupting malicious uses of AI: October 2025

OpenAI has released its latest roundup detailing how the company is finding and shutting down malicious uses of its models, part of a recurring series that's become something like a quarterly report card on AI misuse. The October 2025 edition covers the usual suspects: bad actors trying to bend ChatGPT and related tools toward scams, influence operations, and other real-world harms that OpenAI's policy and safety teams say they've detected and disrupted.

What's notable isn't any single case here, but the pattern these reports have settled into. OpenAI isn't just writing terms of service and hoping people follow them. The company is actively hunting for misuse, banning accounts, and then, months later, publishing what it found. That's a different posture than most tech platforms take, and it's one OpenAI has leaned into hard since these disruption reports started rolling out earlier in the year.

The framing matters too. By calling these efforts to "protect users from real-world harms," OpenAI is making an implicit argument: that the danger from AI misuse isn't hypothetical or far-off, it's happening now, in ways specific enough to detect and document. Scammers, propagandists, and other opportunists are already treating large language models as just another tool in the kit, alongside fake websites and phishing emails.

Whether this transparency push is mostly about safety or mostly about optics is a fair question, and probably both. Either way, the fact that OpenAI keeps putting these reports out, month after month, suggests the company sees ongoing disclosure as part of its licensing to operate at this scale. Enforcement without visibility just isn't going to cut it anymore, not when regulators and the public are watching this closely.