On 13 October 2026, the US Cybersecurity and Infrastructure Security Agency (CISA) warned that attackers are exploiting a zero-day flaw in Ivanti Connect Secure, a virtual private network (VPN) product that companies use to give staff remote access. A zero-day is a flaw attackers use before a fix exists. The bug is tracked as CVE-2026-104935; CVE stands for Common Vulnerabilities and Exposures, the public numbering system for security flaws. CISA says it lets a remote attacker run commands without logging in, and that attackers are using it to steal credentials and data. The Record reports a severity score of 8.6 out of 10, and says federal civilian agencies had until Friday to patch.
How the attack worked
Security firm Volexity found the problem on 5 October while investigating a customer’s network. It traced the break-in to a Connect Secure appliance running version 22.8R2.4, which was fully patched at the time, and says exploitation began on 3 October. Threat-intelligence firm Defused says it saw the first signs on 1 October, according to BleepingComputer; we found that claim only in that one report. Volexity says the attackers chained three weaknesses: a login bypass, a way to run code, and a step that raised their access to full administrator control. Ivanti tracks the first two under CVE-2026-104935 and says a limited number of customers were hit, according to The Record.
Once inside, the attackers went after the keys to the wider network. Volexity says they extracted credentials stored on the appliance, then searched the company’s Active Directory, the directory service that holds staff accounts, for Local Administrator Password Solution (LAPS) passwords and BitLocker recovery keys. They also erased traces from system logs. Volexity could not establish the attackers’ final goal because the activity was detected and disrupted, but judged with high confidence that they were preparing a broader compromise. Its assessment that the motive is criminal rather than espionage carries only low confidence.
What is fixed, and what is not
Ivanti released versions 22.8R2.5, 25.1.3.0, 25.1.3.1 and 25.1.1.2 on 13 October, BleepingComputer reports. Of the three flaws Volexity found, Ivanti patched two that day, according to The Record, and is still working on a fix for the third, which lets attackers gain full control. Volexity says it is due in the coming weeks. There is a further catch: CISA warned that Ivanti’s Integrity Checker Tool may miss this attack because the attacker’s commands run in memory, so confirming a compromise takes network and memory forensics.
For appliances found to be compromised, CISA advises backing up the configuration, performing a factory reset, upgrading to the latest version and restoring the saved configuration. It also says to reset passwords for all accounts, including domain, administrator and local accounts, revoke and reissue stored certificates, and reset stored API (application programming interface) keys and license server credentials.
Did artificial intelligence help?
Volexity assesses with medium confidence that the attacker used large language models (LLMs), the artificial intelligence (AI) systems behind chatbots, to develop the exploit and carry out the later activity. Its reasoning is the complexity of the chain, the habit of sending commands through the exploit itself rather than a simpler backdoor, and scripts built around specific hard-coded credentials instead of generic tools. That is an assessment, not proof. Volexity also says AI models from OpenAI cut its own analysis time substantially.
Our own analysis: patching is not enough
This section is our own reading, not established fact. The Record notes this is the sixth Ivanti flaw added to CISA’s exploited list this year, and here the attackers got in through a device that was already fully patched. That suggests an internet-facing VPN needs monitoring and a plan for stolen passwords, not just updates. Shadowserver, an internet-monitoring group, counts more than 5,600 internet-exposed hosts that appear to be Connect Secure systems worldwide, though BleepingComputer notes it is unknown how many are secured or are decoys. We found no India-specific advisory or count, so Indian organisations using the product should simply follow the vendor and CISA guidance above.