HabrAugust 8, 2026🇷🇺Translated from Russian

RCE Vulnerability in AI Code Editors Cursor, VS Code and Google Antigravity Threatens 50 Million Developers

Researchers at AISLE have disclosed a critical remote code execution vulnerability that affected the popular AI code editors Cursor, Microsoft Visual Studio Code and Google Antigravity. The flaw enabled an attacker to execute arbitrary code on a developer’s machine simply by embedding a malicious link inside a Git commit message. Approximately 50 million users were potentially at risk before patches were issued by all three vendors.

How the attack worked

The exploitation scenario closely mirrored normal development workflows. An attacker only needed to place a crafted link in a commit message. When a developer viewed the commit history inside the IDE and clicked the link, arbitrary code executed with the privileges of the current user. No pop-ups, confirmation dialogs or other indicators appeared during execution.

Successful exploitation allowed an attacker to:

  • Steal API keys for services such as OpenAI, Anthropic and Stripe
  • Install persistent malware that survived editor restarts
  • Read or delete files on the local filesystem
  • Run any commands under the logged-in user account

Why multiple editors were affected

Cursor was built as a fork of Visual Studio Code and inherited its architecture. The same underlying issue was later found in Google Antigravity, which also relies on VS Code components. This demonstrates the “shared component vulnerability” effect: when many AI tools reuse the same foundation, a single architectural flaw can propagate across multiple widely used products.

Why developers are high-value targets

Developer workstations typically contain corporate Git repositories, API and SSH keys, Kubernetes credentials and environment secrets. Compromising one machine often provides an initial foothold into broader corporate infrastructure. AI editors increase this risk by concentrating even more sensitive data and automation capabilities in a single environment.

Practical recommendations

Organizations using the affected editors should:

  • Update to the latest patched versions immediately
  • Scan developer workstations for suspicious activity
  • Rotate any API keys, tokens or credentials that may have been stored locally
  • Treat repository contents, including commit messages, as potentially untrusted
  • Review AI-generated code and actions with heightened scrutiny

Related articles

HabrVulnerabilities & Exploits

How Modern API Attacks Abuse Legitimate Functionality Instead of Exploiting Vulnerabilities

Traditional API incidents once centered on authorization errors, access control issues, SQL injections, and known vulnerability exploitation. Today, many attacks on APIs no longer require finding code flaws; attackers simply use documented methods, valid authorization, and correct parameters at unexpected scale or timing. NGENIX security teams observe three main behavioral patterns: Burst attacks that overload resource-heavy endpoints, Shortwave attacks that exploit race conditions through timed parallel requests, and Carpet Bombing that distributes activity across dozens of endpoints for reconnaissance or scalping. These techniques often bypass WAF because each individual request appears fully compliant with API specifications. Protection is shifting toward behavioral analysis, sliding-window rate limiting, and monitoring of overall client behavior rather than single-request signatures. Real-world cases include sudden catalog scraping during contests and mass reservation of airline seats without purchases, leading to degraded service for legitimate users.

Security NEXTVulnerabilities & Exploits

Critical Vulnerabilities Patched in WHMCS Billing Software for Hosting Providers

WebPros International has disclosed two serious vulnerabilities in its WHMCS billing management platform used by hosting and cloud service providers. CVE-2026-67399 allows unauthenticated remote code execution through unsafe deserialization of untrusted data under specific conditions, potentially compromising the entire server environment and associated data. CVE-2026-67398 affects the 2CheckOut payment gateway module and stems from missing authorization checks, enabling attackers to retrieve sensitive customer information including names, addresses, emails, and phone numbers without authentication. HackerOne assigned CVSS v4.0 scores of 9.3 (Critical) to the first issue and 8.2 (High) to the second. WebPros released fixed versions WHMCS 9.0.8 and 8.13.7, and recommended disabling the 2CheckOut module as a temporary mitigation for the second flaw.

Security NEXTVulnerabilities & Exploits

Cisco Releases Critical Patches for Exploited SQL Injection Flaw in Secure Email Gateway

Cisco Systems has issued security updates for Cisco Secure Email Gateway to address a critical SQL injection vulnerability tracked as CVE-2026-76461. The flaw stems from insufficient input validation during email parsing and allows unauthenticated remote attackers to execute arbitrary SQL commands. Successful exploitation can lead to root-level access on the underlying operating system, enabling full command execution. The vulnerability carries a CVSSv3.1 base score of 9.8 and is rated Critical. Cisco confirmed active exploitation of the issue in September 2026. Recommended fixes include upgrading to versions 16.5.0-780, 16.0.4-3021, or 15.5.5-0141, with strong preference given to the newest release.

AntiMalwareVulnerabilities & Exploits

Telegram Desktop HTML Export Flaw Allowed Stealthy JavaScript Injection into Chat History

Researchers at ExPatch identified a vulnerability in Telegram Desktop that enabled attackers to embed malicious JavaScript into exported HTML chat histories without user detection. The flaw stemmed from insufficient sanitization of button captions added by bots, allowing hidden scripts to execute when the HTML file was opened in a browser. Malicious messages could be forwarded into chats and remain dormant until export, potentially exfiltrating messages, sender names, and timestamps to attacker servers. The issue affected versions 4.15.1 through 6.9.3, with fixes released in beta 6.9.4 and stable version 7.0.1 on July 14. No in-the-wild exploitation was observed, though the attack required specific conditions including an unpatched export and JavaScript-enabled browser. Users are advised to re-export chats after updating or open old files with JavaScript disabled.