Malware Uses AI Model Voting for Attack Decisions
Researchers from Cisco Talos have identified a Windows malware, named CLOSEDQUORUM, that uses up to four AI models to collectively decide its next action instead of receiving commands from a traditional remote server. The AI models can vote to steal credentials, inject code, or persist on a machine, with outcomes and data sent to a Discord channel. The publicly available version lacks working API keys and cannot operate fully, according to Talos.
- Malware requests actions from up to four commercial AI models
- AI models select actions such as steal, inject, persist, or move
- Decisions and stolen data are sent to a Discord channel
- Public version lacks functional keys and cannot connect to AI models
- Researchers say it's the first public case of AI-voted Windows malware
Sources covering this
In this story
More in Cybersecurity
Security startup Hacktron exploits vulnerability in OpenAI forum
Hacktron, an AI cybersecurity startup, found and exploited a security flaw in OpenAI's community forum that allowed access to an employee's internal account using Anthropic's Claude. The breach relied on an image processing vulnerability and was executed in less than 72 hours. Hacktron stopped short of accessing internal code and reported the issue to OpenAI, which then revoked affected tokens. Hacktron received a $6,500 bounty for the discovery.
Microsoft dismantles AI-driven EvilTokens cybercrime platform
Microsoft and partners have taken down EvilTokens, an AI-powered cybercrime platform allegedly used by about 1,000 cybercriminals to compromise over 12,000 Microsoft user email inboxes worldwide. On September 15, Microsoft seized 50 EvilTokens-operated websites and disabled more than 175 related domains. Authorities have arrested two suspects in London accused of fraud and money laundering linked to the platform. At least $1.7 million in losses are tied to EvilTokens.
Hackers claim FBI breach and theft of agent data
A hacking group known as ShinyHunters claims to have breached FBI systems and accessed personal data on thousands of FBI agents and job applicants, including names, home addresses, and phone numbers. Both TechCrunch and 404 Media reviewed samples provided by the hackers and confirmed that some details matched public records. The FBI jobs site and special agent applicant portal were down as of publication. The FBI declined to comment.
Microsoft releases record number of security patches
Microsoft released updates to fix at least 974 security vulnerabilities across Windows and other products, its largest patch release to date. The update includes critical flaws, some already being exploited, and over 100 rated as dangers that could let attackers seize control of systems. The unprecedented volume is partly due to the use of AI for vulnerability detection, but experts warn that rapid growth complicates efforts to test and deploy fixes.