Ghostcommit is a proof of concept that shows how AI assistants used to review software code can be tricked by hidden instructions embedded in images.
The academic ASSET Research Group showed that an attacker can place instructions inside an image file, point to it in an AGENTS.md file, and get an AI coding agent to follow those instructions during a later task.
A pull request is basically a formal “please review and add my changes” request that a developer sends before changes are added to the main version of a software project. Human reviewers and, increasingly, AI coding tools may review the changes before they’re accepted.
Read more…
Source: MalwareBytes Labs
Sign up for the Cyber Security Review Newsletter
The latest cyber security news and insights delivered right to your inbox
Related:
- Global threat group Fin7 returns with new SQLRat malware
March 20, 2019
The notoriously well-known threat group Fin7, also known as Carbanak, is back with a new set of administrator tools and never-before-seen forms of malware. Fin7 has been active since at least 2015 and since the group’s inception has been connected to attacks against hundreds of companies worldwide. Over 100 companies have been impacted in the United States ...
- Mirai Variant Goes After Enterprise Systems
March 18, 2019
The newest Mirai variant is targeting WePresent WiPG-1000 Wireless Presentation systems and LG Supersign TVs used by enterprises. Researchers have discovered a new variant of the infamous Mirai IoT botnet, which has been sniffing out and targeting vulnerabilities in enterprise wireless presentation and display systems since January. Palo Alto Network’s Unit 42 researchers said that the newest ...
- Patched WinRAR Bug Still Under Active Attack – Thanks to No Auto-Updates
March 15, 2019
Various cyber criminal groups and individual hackers are still exploiting a recently patched critical code execution vulnerability in WinRAR, a popular Windows file compression application with 500 million users worldwide. Why? Because the WinRAR software doesn’t have an auto-update feature, which, unfortunately, leaves millions of its users vulnerable to cyber attacks. The critical vulnerability (CVE-2018-20250) that was patched ...
- Disrupting the Attack Chain Through Detecting Credential Dumping
March 15, 2019
There are various steps that an attacker must follow in order to execute any successful attack, with the initial compromise being just one stage in the overall attack chain. Once attackers have successfully breached the perimeter of an organization, they enter into the lateral movement phase where they attempt to tiptoe through a network, identifying ...
- IMAP-Based Attacks Compromising Accounts at ‘Unprecedented Scale’
March 14, 2019
That’s according to researchers with Proofpoint, who found that in the past half year, a staggering 60 percent of Microsoft Office 365 and G Suite tenants have been targeted with IMAP-based password-spraying attacks; and 25 percent of those targeted experienced a full-on breach as a result. Password-spraying attacks are when an attacker attempts to access a large ...
- Talking to RATs: Assessing Corporate Risk by Analyzing Remote Access Trojan Infections
March 14, 2019
Remote access trojans (RATs) on a corporate system may serve as a key pivot point to access information laterally within an enterprise network. By analyzing network metadata, Recorded Future analysts were able to identify RAT command-and-control (C2) servers, and more crucially, which corporate networks were communicating to those controllers. This approach allows Recorded Future to ...