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:
- Adaptavist Group breach spawns imposter emails as ransomware crew claims mega-haul
April 21, 2026
UK enterprise software consultancy The Adaptavist Group is investigating a security breach after an intruder logged in with stolen credentials, while a ransomware crew claims it grabbed far more than the company is currently admitting. In a letter to customers, Adaptavist’s CEO Simon Haighton-Williams said the biz detected an “IT security incident” in late March after ...
- Amtrak data breach exposes 2.1M records, reports suggest larger leak
April 20, 2026
Booking a train ticket shouldn’t come with a side of data exposure, but that’s the situation Amtrak customers are now facing. The rail service is dealing with a breach after hackers claimed to have accessed and released millions of customer records online. The exposed dataset was confirmed to contain at least 2.1 million unique accounts, although ...
- North Korean hackers blamed for $290M crypto theft
April 20, 2026
Over the weekend, hackers stole more than $290 million in cryptocurrency from Kelp DAO, a protocol that allows users to earn yields on idle crypto investments. By Monday, LayerZero, one of the projects affected by the hack, accused North Korea of carrying out the heist. The hack is now the largest crypto theft of the year ...
- Mythos: An AI tool too powerful for public release
April 20, 2026
Anthropic’s most capable model to date, Claude Mythos Preview (aka Mythos), has been described as a “step change” in AI performance, especially on cybersecurity tasks. Anthropic tried to keep Mythos a secret until a few weeks ago, when a data leak revealed the existence of what the company said was its most powerful artificial intelligence to ...
- Hackers are abusing Apple account notifications to distribute malware, steal money and data
April 20, 2026
Scammers have found a way to abuse Apple’s email notification system to deliver phishing messages and trick people into giving away sensitive data and system access. Recently, people started receiving emails from the email.apple.com domain, notifying them of a $899 iPhone purchase via PayPal. The email also shared a phone number for the victims to call, ...
- NIST changes enrichment process for National Vulnerability Database due to surge in CVE submissions
April 20, 2026
The number of reported vulnerabilities has surged so sharply that it forced the National Institute of Standards and Technology (NIST) to change how it ‘enriches’ each entry. Until now, NIST would take a basic CVE record and add structured analysis, to make it more useful in the National Vulnerability Database (NVD). That usually includes severity scoring ...