See also: History of steganography
Polyglot programs have been crafted as challenges and curios in hacker culture since at least the early 1990s. A notable early example, named simply polyglot was published on the Usenet group rec.puzzles in 1991, supporting eight languages, though this was inspired by even earlier programs.5 In 2000, a polyglot program was named a winner in the International Obfuscated C Code Contest.6
In the 21st century, polyglot programs and files gained attention as a covert channel mechanism for propagation of malware.7 8 Polyglot files have practical applications in compatibility.9
A polyglot is composed by combining syntax from two or more different formats, leveraging various syntactic constructs that are either common between the formats, or constructs that are language specific but carrying different meaning in each language. A file is a valid polyglot if it can be successfully interpreted by multiple interpreting programs. For example, a PDF-Zip polyglot might be opened as both a valid PDF document and decompressed as a valid zip archive. To maintain validity across interpreting programs, one must ensure that constructs specific to one interpreter are not interpreted by another, and vice versa. This is often accomplished by hiding language-specific constructs in segments interpreted as comments or plain text of the other format.10
Two commonly used techniques for constructing a polyglot program are to make use of languages that use different characters for comments, and to redefine various tokens as others in different languages. These are demonstrated in this public domain polyglot written in ANSI C, PHP and bash:
Highlighted for Bash
Highlighted for PHP
Highlighted for C
Note the following:
The following is written simultaneously in SNOBOL4, Win32Forth, PureBasicv4.x, and REBOL:
Highlighted for SNOBOL
Highlighted for Forth
Highlighted for BASIC
Highlighted for REBOL
The following file runs as a DOS batch file, then re-runs itself in Perl:
Highlighted for DOS batch
Highlighted for Perl
This allows creating Perl scripts that can be run on DOS systems with minimal effort. Note that there is no requirement for a file to perform exactly the same function in the different interpreters.
Polyglot types include:11
Polyglot markup has been proposed as a useful combination of the benefits of HTML5 and XHTML.12 Such documents can be parsed as either HTML (which is SGML-compatible) or XML, and will produce the same DOM structure either way. For example, in order for an HTML5 document to meet these criteria, the two requirements are that it must have an HTML5 doctype, and be written in well-formed XHTML. The same document can then be served as either HTML or XHTML, depending on browser support and MIME type.
As expressed by the html-polyglot recommendation,13 to write a polyglot HTML5 document, the following key points should be observed:
The most basic possible polyglot markup document would therefore look like this:14
In a polyglot markup document non-void elements (such as script, p, div) cannot be self-closing even if they are empty, as this is not valid HTML.15 For example, to add an empty textarea to a page, one cannot use <textarea/>, but has to use <textarea></textarea> instead.
The DICOM medical imaging format was designed to allow polyglotting with TIFF files, allowing efficient storage of the same image data in a file that can be interpreted by either DICOM or TIFF viewers.16
The Python 2 and Python 3 programming languages were not designed to be compatible with each other, but there is sufficient commonality of syntax that a polyglot Python program can be written than runs in both versions.17
A polyglot of two formats may steganographically compose a malicious payload within an ostensibly benign and widely accepted wrapper format, such as a JPEG file that allows arbitrary data in its comment field. A vulnerable JPEG renderer could then be coerced into executing the payload, handing control to the attacker. The mismatch between what the interpreting program expects, and what the file actually contains, is the root cause of the vulnerability.18
SQL Injection is a trivial form of polyglot, where a server naively expects user-controlled input to conform to a certain constraint, but the user supplies syntax which is interpreted as SQL code.
Note that in a security context, there is no requirement for a polyglot file to be strictly valid in multiple formats; it is sufficient for the file to trigger unintended behaviour when being interpreted by its primary interpreter.
Highly flexible or extensible file formats have greater scope for polyglotting, and therefore more tightly constrained interpretation offers some mitigation against attacks using polyglot techniques. For example, the PDF file format requires that the magic number %PDF appears at byte offset zero, but many PDF interpreters waive this constraint and accept the file as valid PDF as long as the string appears within the first 1024 bytes. This creates a window of opportunity for polyglot PDF files to smuggle non-PDF content in the header of the file.19 The PDF format has been described as "diverse and vague", and due to significantly varying behaviour between different PDF parsing engines, it is possible to create a PDF-PDF polyglot that renders as two entirely different documents in two different PDF readers.20
Detecting malware concealed within polyglot files requires more sophisticated analysis than relying on file-type identification utilities such as file. In 2019, an evaluation of commercial anti-malware software determined that several such packages were unable to detect any of the polyglot malware under test.2122
In 2019, the DICOM medical imaging file format was found to be vulnerable to malware injection using a PE-DICOM polyglot technique.23 The polyglot nature of the attack, combined with regulatory considerations, led to disinfection complications: because "the malware is essentially fused to legitimate imaging files", "incident response teams and A/V software cannot delete the malware file as it contains protected patient health information".24
A Graphics Interchange Format Java Archives (GIFAR) is a polyglot file that is simultaneously in the GIF and JAR file format.25 This technique can be used to exploit security vulnerabilities, for example through uploading a GIFAR to a website that allows image uploading (as it is a valid GIF file), and then causing the Java portion of the GIFAR to be executed as though it were part of the website's intended code, being delivered to the browser from the same origin.26 Java was patched in JRE 6 Update 11, with a CVE published in December 2008.2728
GIFARs are possible because GIF images store their header in the beginning of the file, and JAR files (as with any ZIP archive-based format) store their data at the end.29
Jonas Magazinius; Billy K. Rios; Andrei Sabelfeld (4 November 2013). "Polyglots". Proceedings of the 2013 ACM SIGSAC conference on Computer & communications security - CCS '13. pp. 753–764. doi:10.1145/2508859.2516685. ISBN 9781450324779. S2CID 16516484. Archived from the original on 5 September 2022. Retrieved 5 September 2022. 9781450324779 ↩
Bridges, Robert A.; Oesch, Sean; Verma, Miki E.; Iannacone, Michael D.; Huffer, Kelly M. T.; Jewell, Brian; Nichols, Jeff A.; Weber, Brian; Beaver, Justin M.; Smith, Jared M.; Scofield, Daniel; Miles, Craig; Plummer, Thomas; Daniell, Mark; Tall, Anne M. (2023). "Beyond the Hype: An Evaluation of Commercially Available Machine-Learning-Based Malware Detectors". Digital Threats: Research and Practice. 4 (2): 1–22. arXiv:2012.09214. doi:10.1145/3567432. S2CID 247218744. /wiki/ArXiv_(identifier) ↩
Koch, Luke; Oesch, Sean; Adkisson, Mary; Erwin, Sam; Weber, Brian; Chaulagain, Amul (2022). "Toward the Detection of Polyglot Files". arXiv:2203.07561 [cs.CR]. /wiki/ArXiv_(identifier) ↩
"Benefits of polyglot XHTML5". Archived from the original on 12 October 2011. Retrieved 4 September 2022. https://web.archive.org/web/20111012190748/http://xmlplease.com/xhtml/xhtml5polyglot/ ↩
"Polyglot: A program in eight languages". Archived from the original on 6 September 2022. Retrieved 6 September 2022. https://ideology.com.au/polyglot/ ↩
"15th International Obfuscated C Code Contest (2000)". Archived from the original on 6 September 2022. Retrieved 6 September 2022. https://www.ioccc.org/years.html#2000_tomx ↩
Koch, Luke; Oesch, Sean; Chaulagain, Amul; Dixon, Jared; Dixon, Matthew; Huettal, Mike; Sadovnik, Amir; Watson, Cory; Weber, Brian; Hartman, Jacob; Patulski, Richard (2024). "On the Abuse and Detection of Polyglot Files". arXiv:2407.01529 [cs.CR]. /wiki/ArXiv_(identifier) ↩
"Polyglot Markup: A robust profile of the HTML5 vocabulary". Archived from the original on 9 August 2022. Retrieved 4 September 2022. https://www.w3.org/TR/html-polyglot/ ↩
Polyglot Markup: HTML-Compatible XHTML Documents: 6.4 Void Elements Archived 2 October 2012 at the Wayback Machine. W3C Editor's Draft 9 July 2012. http://dev.w3.org/html5/html-xhtml-author-guide/html-xhtml-authoring-guide.html#empty-elements ↩
"DICOM-TIFF dual personality files". Archived from the original on 5 September 2022. Retrieved 5 September 2022. https://learn.canceridc.dev/dicom/dicom-tiff-dual-personality-files ↩
Schofield, Ed. "Cheat Sheet: Writing Python 2-3 compatible code". Archived from the original on 6 September 2022. Retrieved 6 September 2022. https://python-future.org/compatible_idioms.html ↩
Wolf, Julia (9 February 2011). "OMG WTF PDF". 27th Chaos Communication Congress. Archived from the original on 9 October 2022. Retrieved 6 September 2022. https://fahrplan.events.ccc.de/congress/2010/Fahrplan/events/4221.en.html ↩
Desjardins, Benoit; Mirsky, Yisroel; Ortiz, Markel Picado; Glozman, Zeev; Tarbox, Lawrence; Horn, Robert; Horii, Steven C. (April 2020). "DICOM Images Have Been Hacked! Now What?". American Journal of Roentgenology. 214 (4): 727–735. doi:10.2214/AJR.19.21958. PMID 31770023. S2CID 208318324. Archived from the original on 5 September 2022. Retrieved 5 September 2022. https://www.ajronline.org/doi/10.2214/AJR.19.21958 ↩
"Ubiquitous Bug Allows HIPAA-Protected Malware to Hide Behind Medical Images". 17 April 2019. Archived from the original on 5 September 2022. Retrieved 5 September 2022. https://threatpost.com/hipaa-protected-malware-medical-images/143890/ ↩
Byrd, Christopher. "How to Create a GIFAR". Archived from the original on 6 March 2023. Retrieved 6 March 2023. https://www.riosec.com/articles/how-to-create-a-gifar ↩
Eckel, Benjamin (5 August 2008). "The GIFAR Image Vulnerability". Hackaday. Archived from the original on 6 March 2023. Retrieved 6 March 2023. http://hackaday.com/2008/08/04/the-gifar-image-vulnerability ↩
"CVE-2008-5343". cve.mitre.org. 4 December 2008. Archived from the original on 20 April 2021. Retrieved 20 April 2021. https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2008-5343 ↩
McMillan, Robert (1 August 2008). "A photo that can steal your online credentials". Infoworld.com. Archived from the original on 18 September 2020. https://web.archive.org/web/20200918084425/https://www.infoworld.com/article/2653025/a-photo-that-can-steal-your-online-credentials.html ↩
Rios, Billy (17 December 2008). "Billy (BK) Rios » SUN Fixes GIFARs". Archived from the original on 14 March 2016. Retrieved 20 April 2021. https://web.archive.org/web/20160314083259/http://xs-sniper.com/blog/2008/12/17/sun-fixes-gifars/ ↩
Fjeldberg, Hans (2008). Polyglot Programming - A Business Perspective (PDF) (M.Sc). Norwegian University of Science and Technology. Archived (PDF) from the original on 4 March 2016. Retrieved 28 May 2015. http://theuntitledblog.com/wp-content/uploads/2008/08/polyglot_programming-a_business_perspective.pdf ↩
Gupta, Tripta (19 December 2018). "Analyzing Polyglot Microservices". Medium. Archived from the original on 5 August 2019. Retrieved 5 August 2019. https://medium.com/capital-one-tech/analyzing-polyglot-microservices-f6f159a1a3e7 ↩