The U.S. Secret Service shut down a network that could have paralyzed New York
The U.S. Secret Service shut down a network that could have paralyzed New York#### What if someone told you that a network of 300 SIM servers could paralyze New York City? What if attackers were exploiting critical Cisco vulnerabilities to control corporate networks?
Shortly before the start of the UN General Assembly, the US Secret Service uncovered and dismantled a network of over 300 SIM servers and more than 100,000 SIM cards in the New York area. This network enabled encrypted, anonymous communication and had the capacity to disable New York’s mobile network, paralyze telecommunications, and endanger the city’s crisis infrastructure. This could have included the ability to block emergency calls or prevent the coordination of rescue efforts.
According to security officials, the network was set up in five locations within 35 miles of New York City. The SIM servers could generate tens of millions of text messages in a matter of minutes, overwhelming the critical infrastructure of mobile operators in the event of an attack.
Analysts determined that the network was used for typical smishing and mass spam distribution, but it was also connected to criminal activities involving domestic and foreign state actors. During the investigation, connections were revealed to so-called swatting attacks, deceptive threats that aim to trigger a massive response from law enforcement.
The takedown of this network is unprecedented, not only because of its size and technological sophistication, but also because of the real possibility of paralyzing a city of millions.
Hackers took control of corporate networks via Cisco.
Three critical vulnerabilities were published in Cisco network products: CVE-2025–20352, CVE-2025–20333, and CVE-2025–20362. These vulnerabilities primarily affect enterprise firewalls (Cisco ASA/FTD) and routers/switches with IOS/IOS XE.
The first vulnerability, CVE-2025–20352, affects the SNMP subsystem in IOS and IOS XE platforms. This bug allows an attacker with basic SNMP login credentials to trigger a denial-of-service attack or remotely execute code as root. Exploitation is possible through both IPv4 and IPv6, and attackers have been actively exploiting it against exposed devices worldwide in recent days.
Two more vulnerabilities, CVE-2025–20333 and CVE-2025–20362, target firewall environments, specifically the WebVPN (SSL VPN) component. CVE-2025–20333 enables attackers to perform a buffer overflow on the web server. This leads to the execution of arbitrary code with root privileges and complete device takeover. Originally, this vulnerability required valid VPN authentication.
However, the key element of the attack is its combination with CVE-2025–20362. This vulnerability enables attackers to bypass authentication mechanisms. Together, these two vulnerabilities bypass authentication and enable complete control over the target device. This type of attack was used in the UAT4356/Storm-1849 campaign, attributed to Chinese APT groups.
Both Cisco and CISA have issued urgent warnings recommending immediate updates to patched versions, as there are no alternative solutions available.
How did attackers exploit AI agents in Salesforce to leak corporate data?
In September, a critical vulnerability called ForcedLeak was discovered in Salesforce’s AgentForce. This vulnerability revealed the vulnerability of today’s AI agents in corporate CRM systems to new types of attacks. The attackers exploited the principle of “indirect prompt injection,” which involves using input data to make the AI agent perform unintended actions. Útočníci dokázali podstrčit škodlivé pokyny do jádra Salesforce prostřednictvím formulářů Web-to-Lead, kde pak došlo k exfiltraci důvěrných dat bez vědomí zaměstnanců.
The Web-to-Lead feature allows companies to collect new contacts, but it did not properly validate the inputs, opening the door for hackers to exploit it. All they had to do was enter a prompt with instructions into the “Description” field to make the AI agent unknowingly export and send data to an external domain.
The attack could be timed — the payload could be “sleeping” for several days. When an employee processed the lead using the AI agent, the hidden prompt activated without any user interaction. The result could be the exfiltration of large volumes of CRM data, including contact information, business opportunities, and even data from connected corporate applications.
Salesforce responded quickly by implementing a whitelist of trusted URLs and strengthening input validation. The incident, which occurred on September 25, 2025, set a significant precedent for Salesforce and the entire field of AI security in enterprise environments.
A new variant of XCSSET: This malware threatens not only Safari but also Firefox.
Previously targeting macOS, the XCSSET malware is back with an enhanced arsenal of features and greater potential for harm. It originally spread primarily through infected Xcode projects, Apple’s development environment, and primarily targeted developers. The new variant, however, shows that attacks are becoming increasingly sophisticated and dangerous.
XCSSET now steals sensitive data from Safari, Chrome, and Firefox. Using a modified tool called HackBrowserData, it steals passwords, browsing history, cookies, and payment card information.
The malware continuously monitors the clipboard and, as soon as it detects a cryptocurrency address (e.g., Bitcoin or Ethereum), quickly replaces it with its own address. The user then unwittingly sends the cryptocurrency to the attacker.
The malware uses additional methods to survive a system restart and evade detection. These methods include installing LaunchDaemon, masking via a fake System Settings application, and thoroughly encrypting and obfuscating AppleScripts.
An infected project in the Xcode development environment can spread automatically further within the team and ultimately to end users. This gives the attacker access to valuable data and finances.