Malware Evasion Made Easier With AI, but Behavior Still Gives It Away
CSO, Friday, October 16th, 2026
ReversingLabs flagged Midnight Blizzard's AI-rebuilt malware by behavior weeks before Anthropic's report.
Anthropic reported that Midnight Blizzard used AI agents to automatically rebuild its malware whenever security tools detected it, raising fears of polymorphic AI malware.
ReversingLabs says its behavior-based threat hunting flagged the campaign's Go backdoor PowerShell stager as malicious on August 1, weeks before Anthropic's September 10 report.
It then tracked five versions of the stager over 25 days.
The research suggests that while AI makes code-level evasion easier, behavioral detection still catches the underlying activity.