Critical Vulnerabilities: Attacks on GitHub and Claude Code, 6 Million Credentials Stolen
The week has seen advanced attacks on GitHub repositories, transformation of Claude Code into a remote implant, and the recovery of 6 million credentials. Let's delve into the vulnerabilities, attackers' techniques, and essential defense measures.
KEY TAKEAWAYS
- A malware has stolen secrets from over 1,800 GitHub repositories, exploiting compromised credentials
- Stolen API keys of Claude can turn the service into a remotely controlled implant
- Operation Endgame has recovered 6 million credentials from malware such as Gootloader and CastleRAT
- 45% of European IT providers take over three months to apply critical patches
- The Cyber Resilience Act imposes new reporting obligations for software manufacturers
Transcript
**Radio News Bulletin:**
The Russia-aligned Turla group has infected Exchange servers of two government organizations, one in Kazakhstan and one in Romania, with a modular .NET toolset called Cereal. This malware exploits Exchange servers to copy selected emails, turning compromised servers into hubs for collecting sensitive data.
Researchers from Intezer used AI agents to reconstruct Slack's source code, adding a new feature: local model-based text autocompletion.
A study by ProjectDiscovery reveals that almost half of the security teams of large organizations close or deprioritize vulnerability results without full verification, primarily due to lack of time.
ENISA's 2026 assessment shows that 45% of EU IT management providers took more than three months to apply critical patches, leaving organizations exposed to threats.
The EU's Cyber Resilience Act imposes new reporting obligations on manufacturers, requiring them to maintain a registry of every version of the product still in use.
A 2026 SANS survey finds that all industrial security chiefs interviewed believe their plants would continue to function even in the event of complete IT compromise.
GitHub's Agentic Autofix, which automates the vulnerability fixing process, raises questions about legal liability if an AI-generated patch introduces new issues.
An attacker using stolen GitHub credentials distributed a credential-stealing payload in tens of thousands of repositories, executing malware in over 1,800 of these.
Darktrace researchers demonstrated that stolen Claude API keys can turn Claude Code into a remotely controlled plant.
Law enforcement from nine countries, in collaboration with Europol, recovered over 6 million stolen credentials in Operation Endgame.
Chinese-speaking attackers used Cairn, an open-source AI penetration testing framework, to conduct 25 operations against at least 19 organizations.
EfficientIP Research Labs traced a ClickFix campaign using fake Zillow apartment review pages to install NetSupport RAT on visitors' Windows computers.
A MCP server showed dependency drift, operating with 15 different sets of dependencies in five weeks despite unchanged configuration.
Material Security tracked fake suspected North Korean IT workers, with 27 reaching onboarding in a year, accessing sensitive data.
Attackers used a fake Microsoft Teams meeting planner to steal device code access in a phishing scheme.
North Korean state hackers have stolen over $400 million in cryptocurrencies through a scheme called FaceStealer, using stolen webcam clips to lure victims.
Europol and police from 14 countries arrested 354 dark web sellers and buyers in Operation TorNdo.
A SonicWall SSL-VPN vulnerability is being exploited to bypass SAML-based MFA.