Hackers Deploy Bots Tracking React2Shell Exploits

Hackers deployed an automated, AI-augmented campaign abusing React2Shell CVE-2025-55182, a 10.0 CVSS pre-auth remote code execution affecting React Server Components and Next.js, and confirmed over 900 successful compromises. Investigators found an exposed staging host tied to the Bissa scanner that scans millions of targets, validates exploitation, and stages harvested data rather than performing simple opportunistic dumping.

The exposed host contained more than 13,000 files across 150+ directories tied to scanning, victim staging, credential harvesting, and operator workflow. A dedicated CVE-2025-55182 module performed mass probing and automatic confirmation of hits. The attacker toolchain included Claude Code usage and a local AI-control surface called OpenClaw with a model pool that included claude-sonnet-4-6. A Telegram operator endpoint @bissa_scan_bot provided coordination and proof delivery. The campaign heavily tuned post-exploit collection to capture .env files, cloud metadata endpoints, local credential stores, database dumps, and wallet material. Analysts enumerated 400+ env-batch ZIP archives, 30,000+ distinct .env filenames, and 65,000+ archived file entries collected between 10 and 21 April 2026.

Harvested secrets spanned large AI and cloud providers and high-value SaaS tiers. Key categories included:

• •AI platforms and model providers such as Anthropic, OpenAI, Google, Mistral, and HuggingFace
• •Cloud platforms and infrastructure providers including AWS, Azure, Google Cloud, Cloudflare, and DigitalOcean
• •Payment and financial services credentials and keys

This campaign shows a shift from noisy scanning toward integrated, validated exploitation pipelines that combine scanners, human-in-the-loop operator bots, and AI-assisted debugging and orchestration. The use of generative models to interpret scanner logic, debug collection flows, and refine exfiltration increases operational efficiency and accelerates attacker iteration cycles. For defenders, the high success rate against unpatched environments and the focus on secrets tied to AI and cloud services raises systemic risk beyond single-company breaches.

Immediate priorities are patching CVE-2025-55182, rotating all exposed credentials, auditing object storage for unusual env archives, blocking known scanner infrastructure and Telegram control channels, and hunting for evidence of OpenClaw and model-assisted workflows in incident telemetry.