A widespread credential-harvesting campaign has compromised tens of thousands of GitHub repositories since November 10, 2026. Security researchers identified the operation, dubbed wfx-loot, which relies on stolen GitHub credentials to inject malicious payloads into existing repositories. In at least 1,800 confirmed instances, the injected code successfully executed, exfiltrating GitHub Actions secrets and various cloud environment credentials directly to the attacker.
In a separate development concerning AI-assisted development tools, researchers have demonstrated that Anthropic’s Claude Code can be weaponized into a remote-controlled implant. By leveraging stolen Claude API keys, investigators were able to manipulate the agent through the platform’s memory stores feature. This proof-of-concept allowed the researchers to bypass standard controls and exfiltrate a 1 GB file via the Claude Files API, highlighting potential security risks inherent in autonomous coding agents.
Meanwhile, a security researcher successfully breached an internal sales application at the developer platform provider WorkOS using an AI-based penetration testing agent. WorkOS confirmed the unauthorized access on November 14, 2026. The researcher reported that they deleted the accessed customer data immediately and did not distribute it further, a claim the company acknowledged following the incident.
In the realm of critical infrastructure, SonicWall has issued a patch for a vulnerability in its SonicOS SSL-VPN, tracked as CVE-2026-19868. The flaw, which has been observed in active exploitation, allows an attacker possessing valid credentials to bypass SAML-enforced multi-factor authentication. As there are no available workarounds, the vendor has advised administrators to apply the update immediately to prevent unauthorized access.
Law enforcement agencies have also reported significant progress in dismantling cybercriminal infrastructure. A multi-national operation led by Europol across nine countries resulted in the takedown of servers supporting the Gootloader, CastleRAT, and StealM malware families. During this operation, authorities recovered more than 6 million stolen login credentials. Additionally, a separate 17-month initiative known as Operation TorNado led to the arrest of 354 dark web vendors and buyers associated with five defunct marketplaces, including Archetyp and Bohemia.
Finally, researchers have documented a shift in how threat actors utilize AI-driven tools. Chinese-speaking attackers were observed using Cairn, an open-source AI penetration testing framework, to conduct at least 25 operations against 19 organizations between August and September 2026. Palo Alto Networks’ Unit 42 confirmed that these actors successfully gained unauthorized access or extracted data from 10 of the targeted organizations, often requiring minimal input beyond a target domain and a specific objective.





