'Ghostcommit' hides prompt injection in images to fool AI agents, steal secrets

BleepingComputer

July 11, 2026 at 9:03 AM

Overview

Researchers have demonstrated a new technique called 'Ghostcommit' that can hide prompt injection attacks within PNG image files. This method allows attackers to bypass AI code reviewers like CodeRabbit and Bugbot, which do not examine image files. Once the image is processed by a coding agent, it can lead to unauthorized access to sensitive information, such as secrets stored in a repository's .env file. This technique poses a serious risk to software developers and organizations that rely on automated code review tools, as it can result in the exposure of confidential data. The ability to extract secrets and convert them into a readable format makes it a significant concern for data security.

Key Takeaways

Affected Systems: CodeRabbit, Bugbot, software repositories using .env files
Action Required: Developers should ensure that AI code review tools are configured to scan all file types, including images, and consider implementing additional security measures to protect sensitive data in repositories.
Timeline: Newly disclosed

Original Article Summary

A PNG hiding a prompt injection could steal your repo's secrets, researchers demonstrate. The technique, dubbed 'Ghostcommit,' slipped past AI code reviewers CodeRabbit and Bugbot, which never open image files at all, then convinced a coding agent to read a repo's .env and write every secret into the code as a list of numbers. [...]

Impact

CodeRabbit, Bugbot, software repositories using .env files

Exploitation Status

The exploitation status is currently unknown. Monitor vendor advisories and security bulletins for updates.

Timeline

Newly disclosed

Remediation

Developers should ensure that AI code review tools are configured to scan all file types, including images, and consider implementing additional security measures to protect sensitive data in repositories.

Additional Information

This threat intelligence is aggregated from trusted cybersecurity sources. For the most up-to-date information, technical details, and official vendor guidance, please refer to the original article linked below.

Read Original Article

Related Coverage

Critical Zimbra Flaw Could Let Crafted Emails Run Malicious Code in User Sessions

The Hacker News

Zimbra has issued a warning regarding a serious vulnerability in its Classic Web Client that could allow attackers to execute malicious code through specially crafted emails. This vulnerability falls under the category of stored cross-site scripting (XSS) and poses a significant risk as it could enable unauthorized actions within a user's session. While the flaw has not yet been assigned a CVE identifier, Zimbra is urging all customers to implement the necessary updates to mitigate this risk. The potential for arbitrary code execution raises alarms about data security and user safety, making it crucial for affected users to take prompt action. Companies that rely on Zimbra for email services should prioritize applying the updates to protect their systems from potential exploitation.

Jul 11, 2026

Friday Squid Blogging: “Squidbleed” Vulnerability

Schneier on Security

A long-standing vulnerability in the Squid proxy server, known as 'Squidbleed,' has been identified, which can potentially leak sensitive HTTP requests. This bug has been present for 29 years, raising concerns about the security of systems still using affected versions of the Squid software. Administrators of web servers and proxies that rely on Squid need to take immediate action to mitigate any risks associated with this vulnerability. Researchers have flagged the issue as significant, given the age of the flaw and its potential impact on data security. Users of Squid should verify their software versions and apply necessary updates to protect against possible exploitation.

Jul 10, 2026

Update Now: Critical Zimbra Classic Web Client Flaw Could Expose Mailboxes

Security Affairs

Zimbra has issued a warning regarding a serious stored cross-site scripting (XSS) vulnerability in its Classic Web Client, which is commonly used for accessing Zimbra Collaboration. This flaw allows attackers to execute malicious code when users open compromised emails. The company has released version 10.1.19 to address this vulnerability, which currently does not have a CVE ID. Users of the Classic Web Client should update to this latest version as soon as possible to safeguard their mailboxes from potential exploitation. This incident emphasizes the need for prompt software updates to protect sensitive information from cyber threats.

Jul 10, 2026

CISA looks to remedy ailments from big May credential leak

CyberScoop

In May, a significant leak of credentials prompted the Cybersecurity and Infrastructure Security Agency (CISA) to take action. A forensic report released by CISA outlines their plans to enhance protections for sensitive materials and improve the process for researchers to report vulnerabilities within the agency. This incident highlights the need for stronger security measures, especially within government agencies that handle critical infrastructure data. The leak raises concerns about the potential misuse of exposed credentials, which could lead to unauthorized access and other security risks. CISA's proactive steps are essential to prevent similar incidents in the future and to maintain public trust in their operations.

Jul 10, 2026

Injective Labs GitHub Compromise Pushes Wallet-Key-Stealing npm Packages

The Hacker News

Injective Labs' GitHub repository was compromised by unknown attackers who uploaded a malicious package to the npm registry. This malicious version, identified as @injectivelabs/sdk-ts@1.20.21, was designed to steal private keys and mnemonic seed phrases from cryptocurrency wallets. The attackers disguised the harmful code as telemetry functionality, tricking users into believing it was legitimate. This incident poses a significant risk to users of the Injective Labs SDK, as their sensitive information could be at risk of theft. Cryptocurrency users should be cautious and verify the integrity of packages before installation, especially those that seem to come from compromised sources.

Jul 10, 2026

Cybercriminals Flock to Healthcare Businesses as Attacks Surge

darkreading

Cyberattacks on healthcare businesses are on the rise, with a significant increase in attacks targeting service providers. While hospitals and clinics saw a modest growth in cyber incidents during the first half of 2026, the number of attacks on healthcare-related businesses more than doubled. This surge in attacks poses a serious threat to patient data and the overall healthcare infrastructure, making it essential for these organizations to bolster their cybersecurity measures. The rising trend indicates that cybercriminals are increasingly focused on exploiting vulnerabilities in the healthcare sector, which could have dire consequences for patient safety and privacy. As attackers become more aggressive, healthcare providers must remain vigilant and proactive in their defenses.

Jul 10, 2026