OpenAI and Microsoft shut down state-backed hackers

While everyone was busy worrying about autonomous AI wiping out the internet with self-replicating super-viruses, actual nation-state hackers were already using ChatGPT. Their actual use cases, however, look surprisingly mundane.

OpenAI and Microsoft Threat Intelligence teamed up to terminate accounts tied to five state-affiliated cyber groups. The list includes two China-linked actors known as Charcoal Typhoon and Salmon Typhoon, Iran's Crimson Sandstorm, North Korea's Emerald Sleet, and Russia's Forest Blizzard.

Treating frontier AI like a junior intern

Instead of unleashing sci-fi cyber weapons, these sophisticated groups mostly used OpenAI's tools for daily grunt work.

Charcoal Typhoon used the models to research cybersecurity tooling, debug code, and generate scripts. Salmon Typhoon translated technical papers and researched basic methods for hiding system processes.

Iran's Crimson Sandstorm leaned on the platform to assist with web development scripts and draft spear-phishing messages, while North Korea's Emerald Sleet researched public vulnerabilities and identified defense experts across the Asia-Pacific region.

Russia's Forest Blizzard primarily used the models for open-source research into satellite communication protocols, radar imaging technology, and general scripting tasks.

Both OpenAI and Microsoft confirmed that they haven't observed any novel, AI-enabled attack techniques. In practice, attackers are simply using large language models as basic productivity software for tasks they could already accomplish with public search engines and standard utilities.

Drawing the line on nation-state abuse

In response, OpenAI pulled the plug on the identified accounts. The company explained that its internal security teams now use AI models to trace malicious behavior patterns and continuously iterate on safety guardrails.

Microsoft also published a set of core principles for dealing with state-sponsored advanced persistent threats. Whenever Microsoft detects advanced cyber groups abusing its APIs or systems, it will immediately disable their accounts and revoke access.

Crucially, Microsoft also committed to tipping off other AI providers whenever it spots bad actors misusing rival systems, aiming to build a shared defensive network across the tech sector.

Why it matters

This release delivers a much-needed reality check against the wild doom-mongering around AI cyber threats. For everyday users and corporate defenders, current models like GPT-4 do not hand attackers magical zero-day exploits. The immediate risk remains polished phishing emails and faster reconnaissance, which standard security hygiene can still counter.

For the AI industry, this marks the formal beginning of coordinated ecosystem policing. Microsoft leveraging its enterprise threat intelligence to clean up OpenAI's user base creates a blueprint that every serious model provider will have to copy.

That said, simply banning accounts feels like playing whack-a-mole against well-funded adversaries. State-backed groups can easily rotate IP addresses, cycle through fraudulent accounts, or migrate to uncensored open-source models hosted on their own hardware. Cloud-level bans slow them down, but they do not solve the underlying offensive shift.