Even With the Cloud, Client Security Still Matters

As usual, this post does not necessarily represent the views of my employer (past, present, or future).

It’s Friday afternoon and the marketing manager receives an email with the new printed material proofs for the trade show. Double clicking the PDF attachment, his PDF reader promptly crashes.

“Ugh, I’m gonna have to call IT again. I’ll do it Monday morning,” he thinks, and turns off his monitor before heading home for the weekend.

Meanwhile, in a dark room somewhere, a few lines appear on the screen of a laptop:

[*] Sending stage (205891 bytes) to
[*] Meterpreter session 1 opened ( -> at 2017-12-27 16:29:13 -0800
msf exploit(multi/handler) > sessions 1
[*] Starting interaction with 1...
meterpreter > sysinfo
Computer        : INHUMAN-WIN7
OS              : Windows 7 (Build 7601, Service Pack 1).
Architecture    : x64
System Language : en_US
Domain          : ENTERPRISE
Logged On Users : 2
Meterpreter     : x64/windows

Finally, the hacker had a foothold. He started exploring the machine remotely. First, he used Mimikatz to dump the password hashes from the local system. He sent the hashes to his computer with 8 NVidia 1080Ti graphics cards to start cracking, and then kept exploring the filesystem of the marketing manager’s computer. He grabbed the browsing history and saved passwords from the browser, and noticed access to a company directory. He started a script to download the entire contents through the meterpreter session. He started to move on to the network shares when his password cracking rig flashed a new result.

“That was fast,” he thought, looking over at the screen. “SuperS3cr3t isn’t much of a password.” He used the password to log in to the company’s webmail and forwarded the “proofs” (in fact a PDF exploiting a known bug in the PDF reader) to one of the IT staffers with a message asking them to take a look at why it wouldn’t render.

Dissatisfied with waiting until the next week for an IT staffer to open the malicious PDF, he started looking for another option. He began by using his access to a single workstation to look for other computers that were vulnerable to some of the most recent publicly known exploits. Surprisingly, he found two machines that were vulnerable to MS17-010. He sent the exploit through his exisiting meterpreter session and crossed his fingers.

Moments later, he was rewarded with a second Meterpreter session. Looking around, he was quickly disappointed to realize this machine was freshly installed and so would not contain sensitive information or be hosting interesting applications. However, after running Mimikatz again, he discovered that another one of the IT staff had logged into this machine, probably as part of the setup process.

He threw the hashes into his password cracking rig again and started looking for anything else interesting. In a few minutes, he realized this machine was devoid of anything but a basic Windows setup – not even productivity applications had been installed yet. He returned to the original host and looked for anything good, but only found a bunch of marketing materials that were basically public information.

Frustrated, he banged on his keyboard until he remembered the scraped company directory. He went and looked at the directory information for the IT staffer and realized it not only included names and contact informatuon for employees, but also allowed employees to include information about hobbies and interests, plus birthdays and more. He took the data from the IT staffer, split it up into all the included words, and placed it into a wordlist for his password cracking rig. Hoping that would get him somewhere, he went for a Red Bull.

When he came back, he saw another result on his password cracker. This surprised him slightly, because he had expected more of an IT staffer. He was even more surprised when he saw that the password was “Snowboarding2020!” Though it met all the company’s password complexity requirements, it was still an incredibly weak password by modern standards.

Using this new found password, he logged into the workstation belonging to the IT engineer. He dumped the local hashes to look for further pivoting opportunities, but found only the engineer’s own password hash. As he started exploring the filesystem, however, he found many more interesting options. He quickly located an SSH private key and several text files containing AWS API keys. It only took a little bit of investigation to realize that one of the AWS API keys was a root API key for the company’s production environment.

Using the API key, he logged in to the AWS account and quickly identified the virtual machines running the company’s database servers containing user credentials and information. He connected with the API keys he had and started dumping the usernames and password hashes. Given that the hashes were unsalted SHA-1, he figured it shouldn’t take long for his password cracking rig to work through them.

A day later, he was posting an offering for the plaintext credential database for just a fraction of a bitcoin per customer. Satisfied, he started hunting for the next vulnerable enterprise.

While the preceeding story was fiction, it’s an all too-common reality. Many modern enterprises have put considerable effort into hardening their datacenter (be it virtualized or physical) but very little effort into hardening workstations. I often work with companies that seem to believe placing their applications into the cloud is a security panacea. While the cloud offers numerous security benefits – major cloud providers have invested heavily into security, monitor their networks 24/7, and a cloud service is clearly heavily segregated from the corporate network – it does not solve all security problems.

An attacker who is able to compromise a workstation is able to do anything that a legitimate user of that workstation would be able to do. In the example above, the AWS keys stored on a workstation proved critical to gaining access to a treasure trove of user information, but even a lower level of access can be useful to an attacker and dangerous to your company.

The 2017 Verizon DBIR provides data to support this. 66% of malware began with malicious email attachments (client-based), 81% of breaches involved stolen credentials (pivoting), and 43% of attacks involved social engineering (tactics against legitimate users).

Imagine the you have customer service representatives who log in to an application hosted in the cloud to process refunds or perform other services. An attacker with access to a customer service workstation might be able to grab their username and password (or saved cookies from the browser) and then use it to buy expensive items and refund them to themselves. (Or change delivery addresses, issue store credits, or other costly expenditures.)

In a hospital, compromising a workstation used by doctors and nurses would lead, at a minimum, to a major HIPAA breach. In the worst case, it could be used to modify patient records or order medications what could be dangerous or fatal to a patient. Each environment needs to consider the risks posed by the access granted from their workstations and clients.

Attackers will take the easiest route to the data they seek. If you’ve spent some effort on hardening your servers (or applications in the cloud), that may well be through the workstation or client. Consider all entry points in your security strategy.