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:
- Siemens Patches Firewall Flaw That Put Operations at Risk
November 14, 2018
The industrial company on Tuesday released mitigations for eight vulnerabilities overall. Siemens AG on Tuesday issued a slew of fixes addressing eight vulnerabilities spanning its industrial product lines. The most serious of the patched flaws include a cross-site scripting vulnerability in Siemens’ SCALANCE firewall product. The flaw could allow an attacker to gain unauthorized access to ...
- State-Sponsored Actors Focus Attacks on Asia
November 14, 2018
Southeast Asia is the most actively attacked region, accordingly to Cyber Security firm, Group-IB. Their annual Hi-Tech Crime Trends Report 2018 advises, “In just one year, 21 state-sponsored groups were detected in this area, which is more than in the United States and Europe.” Although, not only state-sponsored groups are focusing their attention on this ...
- 7 New Meltdown and Spectre-type CPU Flaws Affect Intel, AMD, ARM CPUs
November 14, 2018
Disclosed earlier this year, potentially dangerous Meltdown and Spectre vulnerabilities that affected a large family of modern processors proven that speculative execution attacks can be exploited in a trivial way to access highly sensitive information. Since then, several more variants of speculative execution attacks have been discovered, including Spectre-NG, SpectreRSB, Spectre 1.1, Spectre1.2, TLBleed, Lazy FP, NetSpectre and Foreshadow, patches for which were released ...
- October 2018’s Most Wanted Malware: For The First Time, Remote Access Trojan Reaches Top 10 Threats
November 13, 2018
For the first time, Check Point threat intelligence researchers found that a remote access Trojan (RAT) has reached the Global Threat Index’s Top 10. Dubbed “FlawedAmmyy”, this type of attack allows attackers to remotely control the victim’s machine– gaining full access to the machine’s camera and microphone, collecting screen grabs, stealing credentials and sensitive files, ...
- Microsoft patches Windows zero-day used by multiple cyber-espionage groups
November 13, 2018
Microsoft released today its monthly roll-up of security patches known as Patch Tuesday. This month, the Redmond-based company has fixed 62 security flaws. Among the 62 fixes, there is also a fix for a zero-day vulnerability that was under active exploitation before today’s patches were made available. The zero-day, tracked as CVE-2018-8589, impacts the Windows Win32k component. Microsoft ...
- Using Machine Learning to Cluster Malicious Network Flows From Gh0st RAT Variants
November 13, 2018
Cybercriminals have become more and more creative and efficient in their efforts to successfully bypass network security. Reports of unauthorized network intrusions that have compromised enterprise security, resources, and data, plague experts on a day-to-day basis, and will continue to do so if not prevented by a more efficient detection system or method. Currently, attackers use polymorphism, ...