cybersecuritynews
'Ghostcommit' hides prompt injection in images to fool AI agents, steal secrets
July 10, 2026
Back to all postsIn a significant cybersecurity development, researchers have discovered a new malware technique known as 'Ghostcommit'. This malicious software hides prompt injection within images, bypassing AI-powered security systems that typically detect such threats. The attackers exploit GitHub's auto-execution feature for scripts in pull requests to gain access to sensitive data.
Source: BleepingComputer
cybersecuritynews
Concerned about this threat to your business?
We help Canadian SMBs deploy the controls discussed in this article. Free 30-minute assessment — no obligation.