Ghostcommit attack hides malicious AI instructions in images


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:

  • Apple fixes actively exploited iOS zero-day on older iPhones, iPads

    January 23, 2023

    Apple has backported security patches addressing a remotely exploitable zero-day vulnerability to older iPhones and iPads. This bug is tracked as CVE-2022-42856, and it stems from a type confusion weakness in Apple’s Webkit web browser browsing engine. Read more… Source: Bleeping Computer  

  • FBI Confirms Lazarus Group Cyber Actors Responsible for Harmony’s Horizon Bridge Currency Theft

    January 23, 2023

    The FBI continues to combat malicious cyber activity, including the threat posed by the Democratic People’s Republic of Korea (DPRK) to the U.S. and our private sector partners. Through our investigation, we were able to confirm that the Lazarus Group (also known as APT38), cyber actors associated with the DPRK, are responsible for the theft ...

  • CISA Adds One Known Exploited Vulnerability to Catalog

    January 23, 2023

    CISA has added one new vulnerability to its Known Exploited Vulnerabilities Catalog, based on evidence of active exploitation. This type of vulnerability is a frequent attack vector for malicious cyber actors and poses a significant risk to the federal enterprise. Note: To view the newly added vulnerabilities in the catalog, click on the arrow in the ...

  • Russia’s largest ISP says 2022 broke all DDoS attack records

    January 23, 2023

    Russia’s largest internet service provider Rostelecom says 2022 was a record year for Distributed denial of service attacks (DDoS) targeting organizations in the country. DDoS attacks are cyberattacks aimed at making an internet-connected website or service unavailable by overwhelming it with many requests that deplete the server’s ability to accept new connections, causing the service to ...

  • Hacker finds copy of TSA no-fly list on exposed cloud storage

    January 22, 2023

    A copy of the U.S. Transportation Security Administration’s “no-fly list” has been found by a Swiss hacker exposed on the open internet in yet another case of misconfigured cloud storage. First reported by The Daily Dot, the exposure of the database was found by a Swiss hacker known as “maia arson crimew” on a server run ...

  • 37 million T-Mobile customers hacked in data breach

    January 20, 2023

    T-Mobile said a “bad actor” accessed personal data from 37 million current customers in a November data breach. In a regulatory filing Thursday, the company said the hacker stole customer data that included names, billing addresses, emails, phone numbers, dates of birth, T-Mobile account numbers, and information describing the kind of service they have with the ...