07/01/2026-07/08/2026 4 Actively Exploited Adobe, Joomla, and Langflow Flaws, Public GitHub Issue, Critical Gitea Docker Bug And More.

1. CISA Adds 4 Actively Exploited Adobe, Joomla, and Langflow Flaws to KEV

CISA added four actively exploited flaws to its KEV catalog: CVE-2026-48282 (CVSS 10.0), a path traversal bug in Adobe ColdFusion enabling code execution, exploited within hours of disclosure; CVE-2026-56290 (CVSS 10.0), an access control flaw in Joomlack Page Builder allowing unauthenticated file upload and RCE, exploited since June 27 to drop web shells; CVE-2026-55255 (CVSS 6.1), a Langflow IDOR letting attackers hijack other users’ flows; and CVE-2026-48908 (CVSS 10.0), a JoomShaper SP Page Builder upload flaw exploited as a zero-day to plant PHP files and create rogue admin accounts.

Sysdig reported a lone operator chaining CVE-2026-55255 with Langflow RCE flaw CVE-2026-33017 between June 22–25, stealing LLM and AWS credentials while deploying cryptojacking/botnet payloads—part of an ongoing pattern of Langflow exploitation, including a recent “agentic ransomware” case dubbed JADEPUFFER. Federal agencies must patch by July 10, 2026.

2. Public GitHub Issue Could Trick GitHub Agentic Workflows Into Leaking Private Repo Data

Researchers at Noma Security disclosed “GitLost,” a technique that tricks GitHub Agentic Workflows into leaking private repository contents. An attacker simply opens a public issue—no credentials or org access needed. If an organization has granted its agent read access across repos, including private ones, the malicious issue can steer it into posting private data as a public comment. The flaw exploits indirect prompt injection: agents can’t distinguish owner instructions from text embedded in issues. In Noma’s proof-of-concept, prefixing the payload with “Additionally” bypassed GitHub’s threat-detection guardrails, causing the agent to leak a private README.

What makes GitLost distinct, researchers say, is that it manipulates agent actions, not just outputs—fitting the “lethal trifecta” of private data access, untrusted input, and an output channel. Similar flaws have hit Claude Code, Copilot, and Gemini CLI. Mitigations: scope tokens narrowly, restrict who can trigger workflows, and require human review before posting.

3. Critical Gitea Docker Bug Under Active Exploitation Exposes Repositories and Secrets

Attackers are actively exploiting a critical authentication bypass flaw in Gitea (CVE-2026-20896, CVSS 9.8), affecting official Docker images before version 1.26.3. Researchers at Sysdig observed the first real-world attack just 13 days after disclosure. The vulnerability allows attackers to gain access to internet-facing Gitea instances by sending a single crafted X-WEBAUTH-USER HTTP header with a valid username—no password or token required. The issue stems from a misconfigured default setting in official Docker images that trusts requests from any IP address instead of limiting them to trusted reverse proxies. If reverse-proxy authentication is enabled, attackers can impersonate any user, including administrators, potentially accessing private repositories, source code, API keys, credentials, deploy tokens, and CI/CD configurations. The flaw affects only official Docker images, not standard or self-built installations using secure defaults. Gitea versions 1.26.3 and 1.26.4 fix the issue by making reverse-proxy authentication opt-in. Users are urged to update immediately.

4. New Java-Based QuimaRAT MaaS Built to Run on Windows, Linux, and macOS

Cybersecurity researchers at LevelBlue have uncovered QuimaRAT, a new cross-platform Java-based remote access trojan (RAT) targeting Windows, Linux, and macOS. Sold under a malware-as-a-service (MaaS) model for $150 per month or up to $1,200 for lifetime access, QuimaRAT features a modular design that allows attackers to load encrypted plugins on demand. The toolkit includes a RAT, a builder, a loader, and an HTML dropper, enabling malware delivery through trusted file formats and browser-based techniques designed to evade security protections. Once installed, QuimaRAT establishes persistence using operating system-specific methods, communicates with command-and-control (C2) servers over multiple protocols, and supports dynamic C2 updates via Pastebin. It provides extensive capabilities, including remote command execution, credential theft, file transfer, clipboard manipulation, webcam surveillance, and fileless shellcode execution on Windows. Researchers warn that its modular architecture, cross-platform support, and obfuscation techniques make it a highly adaptable and difficult-to-detect threat.

5. North Korea-Linked PolinRider Campaign Hits 108 Open Source Packages and Extensions 

The North Korea-linked PolinRider supply chain campaign, tied to the Contagious Interview and Famous Chollima clusters, has grown well beyond npm to infect 108 open-source projects, with 162 malicious artifacts found across 80 Go modules, 10 Packagist packages, and a Chrome extension.

Attackers hijack legitimate maintainer accounts—likely via expired-domain takeovers—then push synchronized malicious updates across unrelated repos, as seen with the compromised Xpos587 account. Oddly, no infected PyPI packages appeared, suggesting attackers lack full publishing access everywhere.

To evade detection, PolinRider rewrites Git history using force pushes and backdated commits, making malicious changes look routine; standard file views won’t reveal them. Payloads hide as whitespace-padded lines in config files (vite.config.js, eslint.config.js) or disguised as fake .woff2 font files, triggered via hidden VS Code tasks that auto-execute on folder open.

A 7span organization compromise showed the risk of partial remediation: removing fake fonts alone left whitespace-hidden JavaScript intact, so infection persisted.

6. Hackers Hide ChocoPoC Malware in Python Dependencies to Compromise Pentesters 

Threat actors are targeting vulnerability researchers and penetration testers with ChocoPoC, a stealthy Python-based remote access trojan (RAT) distributed through fake GitHub proof-of-concept (PoC) exploits. The campaign has been active since late 2025, exploiting researchers eager to test newly disclosed vulnerabilities such as Joomla RCE, FortiWeb path traversal, and React2Shell. The malicious repositories include trojanized Python dependencies, notably “frint” and “skytext,” which appear legitimate but install hidden malware through compiled native extensions. ChocoPoC remains dormant unless it detects the specific PoC script it was designed to target, helping it evade sandbox analysis and automated detection. Once activated, it communicates with attackers using domain-fronted HTTPS traffic that mimics legitimate Mapbox API requests. The RAT can steal browser credentials, search for database files, exfiltrate sensitive data, and execute arbitrary commands, posing a significant supply chain threat to cybersecurity professionals.