- Threat actors exploited the critical SSRF bug in LMDeploy toolkit just 12 hours after its public disclosure.
- The vulnerability, CVE-2026-33626, allows attackers to steal cloud credentials and scan internal networks via the platform’s image loader.
- Sysdig observed the attack using the flaw to target AWS metadata and port-scan local services, highlighting a dangerous trend in AI infrastructure.
- Separate campaigns are actively exploiting WordPress plugin flaws and targeting internet-exposed industrial control systems globally.
Cyber attackers are moving at unprecedented speed, with the first exploitation of a severe security flaw in the LMDeploy AI toolkit occurring less than 13 hours after its vulnerability disclosure. According to cloud security firm Sysdig, the exploitation attempt was detected on April 22, 2026, originating from a specific threat actor’s infrastructure. This incident underscores the immediate risks in emerging AI deployment platforms.
The vulnerability, tracked as CVE-2026-33626 with a CVSS score of 7.5, is a Server-Side Request Forgery (SSRF) flaw in the toolkit’s vision-language module. The project maintainers issued an advisory stating the `load_image()` function fetches arbitrary URLs without validation, allowing access to internal resources. Consequently, successful exploitation can lead to stolen cloud credentials and lateral network movement.
Sysdig said the attacker used the bug as a generic HTTP SSRF primitive in a detailed three-phase attack. The adversary targeted the AWS Instance Metadata Service and Redis before confirming egress with an OOB DNS callback and port-scanning the loopback interface. This rapid weaponization fits a pattern of critical AI infrastructure vulnerabilities being exploited within hours of publication.
Meanwhile, security teams are tracking other widespread campaigns. Threat actors are actively exploiting critical vulnerabilities in two WordPress plugins, Ninja Forms and Breeze Cache, to achieve full site takeover. Simultaneously, a separate global campaign from September to November 2025 targeted internet-exposed Modbus PLCs across 70 countries, as detailed in a report by Cato Networks researchers said. These coordinated efforts demonstrate the expanding attack surface facing both web and industrial systems.
✅ Follow BITNEWSBOT on Telegram, Facebook, LinkedIn, X.com, and Google News for instant updates.
