Russian‑linked GREYVIBE cyber‑espionage group weaponises ChatGPT across the attack chain

What happened

Security researchers at WithSecure have identified a previously unknown threat group, dubbed GREYVIBE, that has been running a sustained espionage campaign against Ukrainian entities since at least August 2025. The group’s hallmark is the systematic use of generative AI tools – OpenAI’s ChatGPT, Google’s Gemini and the image‑generation service Ideogram – to create lures, write malicious code, spin up command‑and‑control (C2) infrastructure and even automate post‑compromise tasks.

Legal basis

The activity breaches multiple international and domestic statutes. Under the EU General Data Protection Regulation (GDPR), the unlawful processing of personal data of EU citizens – which includes many Ukrainian‑based individuals who are also EU residents – triggers Articles 5, 6 and 32, obligating controllers to implement appropriate technical and organisational measures. The United Kingdom’s Data Protection Act 2018, which mirrors GDPR, would treat the same conduct as a serious infringement. In the United States, the California Consumer Privacy Act (CCPA) extends protection to California‑resident data that may have been harvested during the campaign, exposing the perpetrators to civil penalties of up to $7 500 per violation.

Impact on users and companies

• Victims – Military officers, civil servants, journalists and employees of private firms received spear‑phishing emails that mimicked legitimate Ukrainian services. Some messages redirected users to fake CAPTCHA pages, while others advertised a non‑existent adult‑club site. Once a victim clicked, the AI‑generated payload – dubbed LegionRelay – was dropped. The malware’s code, apparently drafted with LLM assistance, contained sloppy naming conventions such as “letsrollboyos” and “cuteuwu,” making it easier for defenders to attribute the infection.

• Data exposure – The compromised systems allowed the attackers to exfiltrate personal identifiers, location data and internal communications. For EU‑based personnel, this constitutes a GDPR breach that could lead to fines of up to €20 million or 4 % of annual turnover, whichever is higher.

• Infrastructure risk – By automating the provisioning of cloud servers and domain registration through AI‑generated scripts, GREYVIBE reduced the time between reconnaissance and deployment. This rapid “turn‑key” approach increases the attack surface for organisations that rely on third‑party services, potentially pulling them into the cross‑fire of sanctions or export‑control violations.

Why AI matters

The researchers describe the group’s use of generative AI as operationally integrated rather than experimental. Large language models can:

1. Draft convincing phishing copy – By feeding a few keywords, the model produces culturally and linguistically accurate lures that bypass basic spam filters.
2. Write obfuscated code – Prompting the model to “create a PowerShell downloader that hides its network traffic” yields functional snippets that can be stitched together with minimal human editing.
3. Generate infrastructure scripts – AI can produce Terraform or CloudFormation templates that stand up servers, DNS records and SSL certificates in minutes.
4. Assist post‑compromise – Commands for credential dumping, lateral movement and data exfiltration can be generated on the fly, allowing low‑skill operators to execute complex techniques.

Regulatory implications

The campaign highlights a gap in current cyber‑security regulations. While GDPR and CCPA impose heavy penalties for data breaches, they do not directly address the use of AI in the creation of malicious tools. Some jurisdictions are beginning to close that gap:

• The European Commission’s proposed AI Act classifies “generative AI systems used for illicit purposes” as high‑risk, mandating transparency logs and mandatory reporting of misuse.
• The United States is considering amendments to the Cybersecurity Information Sharing Act (CISA) that would require vendors to disclose when AI‑generated code is incorporated into malware samples shared with Information Sharing and Analysis Centers (ISACs).

What changes are needed

1. Mandatory AI‑use disclosure – Companies that develop security products should label any AI‑generated components in threat‑intel feeds, enabling analysts to track the diffusion of LLM‑crafted malware.
2. Enhanced supply‑chain vetting – Organisations must audit third‑party code for AI‑generated artefacts, especially when the code originates from public repositories such as GitHub.
3. Training for defenders – Security teams should receive instruction on how to recognise AI‑style writing patterns, including repetitive phrasing and overly generic variable names.
4. Policy updates – Data‑protection authorities should issue guidance on how AI‑assisted breaches affect the calculation of fines, acknowledging that the speed and scale of attacks may increase the number of affected data subjects.

Conclusion

GREYVIBE is not a polished, elite cyber‑espionage outfit; it is a crew that leverages generative AI to compensate for skill gaps and accelerate its operations. The group’s sloppy operational security – uploading malware to public services, leaving behind whimsical file names and exposing backend infrastructure – gave researchers a rare window into its methods. As AI tools become more accessible, the line between a hobbyist hacker and a state‑aligned threat actor will blur, pressuring regulators to adapt existing privacy and cybersecurity frameworks.

For further reading, see WithSecure’s full technical report here.