Invisible Code Attack Baffles Security Defenses

A sophisticated supply-chain attack has emerged, utilizing invisible Unicode characters to hide malicious code within software packages. Researchers recently identified 151 such packages uploaded to GitHub between March 3 and March 9, 2026. This tactic makes traditional security measures, including manual code reviews and static analysis tools, largely ineffective because the harmful code is rendered unreadable to humans.

Security experts suspect that the attack group, dubbed "Glassworm," may be employing AI to generate these deceptively legitimate packages at scale. The visible code in these packages is often high-quality, with subtle changes like documentation updates or minor refactors. The malicious payloads are embedded using special Unicode characters from the Public Use Areas, which computers interpret as executable code but appear as blank space to human reviewers.

This invisible code technique, first observed last year, was initially used to conceal prompts for AI engines. It has since evolved into a method for embedding traditional malware. Once decoded during runtime, these payloads can steal tokens, credentials, and secrets. Similar malicious packages have also been found on npm and the VS Code marketplace, indicating a broader campaign.