Difference between revisions of "CAPTCHA"

Latest revision as of 16:07, 12 February 2023

An example of a CAPTCHA implementation on Wikipedia's website. This particular CAPTCHA is used to ensure that only human users are able to create new Wikipedia accounts.^[7] Image Credit: Wikpedia

CAPTCHAs are designed around the limitations of current artificial intelligence, which struggle with particular tasks that humans generally have no difficulty with.^[3] However, as artificial intelligence technology progresses, new programs are often able to reliably evade existing security measures.^[8][9] Computer scientists view this as a win-win—either a CAPTCHA version cannot be defeated and a site remains secure, or the program employed to evade the existing CAPTCHA has successfully solved an open problem in artificial intelligence and significantly advanced the field.^[1][2] Regardless, CAPTCHAs themselves must continually evolve in order to remain a step ahead of malicious programs and evasive strategies, which has led to the propagation of new challenge systems and versions.^[8][9]

CAPTCHA and its successors have been criticized for presenting a barrier to web accessibility and generating undue challenges for users with disabilities or atypical skill and knowledge sets. Such users, in addition to potentially lacking abilities presumptively possessed by the human user audience, often actively utilize technology to enable or supplement their online experiences (e.g., screen reading or speech recognition software.) Their assistive devices, being programs, are by design regularly unable to appropriately interpret or bypass certain CAPTCHAs in order to verify personhood and access the site.^[10][11]

Evolution of Programs

As technology progresses, computer scientists have seen a burgeoning need to identify the bounds of artificial intelligence and, more specifically, distinguish between humans and computers.^[3] Turing tests and CAPTCHAs have met that need on both a theoretical^[12] and practical^[3] level.^[13]

Turing Test

Alan Turing, in his 1950 paper “Computing Machinery and Intelligence”, first proposed the idea of an imitation test for distinguishing between computers and humans. Turing describes an old parlor game, the Imitation Game, wherein a man and a woman go into a room together, and an interrogator remains in a separate room. The interrogator may pose input questions to the two, and, based solely on their output answers—ideally, typewritten so as to anonymize answers and disguise any effects of handwriting or vocal timbre—must decide which answers belong to the man and which to the woman. Turing proposes that it be modified for use as a metric of computer development and intelligence. Rather than a man and a woman, the interrogator attempts to distinguish between a human and a computer.^[12]

Turing argues that if the functions and outputs of a human and a machine are indistinguishable, then for all intents and purposes the machine is indeed thinking in the way that a human does. He refers back to the solipsistic philosophical notion that the only human one can be truly certain exists or is thinking in the way that they do is oneself, and says that machines are no different in terms of thought.^[12] Consequently, this metric can be used not only to determine whether an unknown entity is a computer or a human, but also as a benchmark of artificial intelligence progression generally.^[12][13] Such imitation games are now commonly (and eponymously) known as Turing Tests, and are widely used within the domains of computer science and artificial intelligence.^[13]

Upside-Down or Reverse Turing Test

The original role of a Turing test was for humans to be able to distinguish between a human and a computer pretending to be a human.^[12][14] As such, CAPTCHAs, which are designed for computers (those running the program) to be able to distinguish between a human and a computer pretending to be a human, have been described as a upside-down Turing test of sorts.^[14] Conversely, a reverse Turing test would perhaps more aptly be described as one wherein a human attempted to convince a computer moderator that they are a computer, although this terminology is contested.^[14]

This role has been criticized as computers exercising undue authority over humans and barring access to sites created for humans by humans. That is, computers are taking on an active role in the digital landscape, rather than remaining tools exercised by human actors, and potentially exerting dominance over humans in a way that will only continue to escalate.^[14] By the most extreme interpretations, artificial general intelligence and artificial super intelligence could pose an existential threat to humanity.^{[14][15][16][17]}

Dual Uses

Digitizing Information

reCAPTCHA (see below) served an innovative dual purpose beyond that of the typical CAPTCHA program. In addition to guarding websites against bots posing as human, reCAPTCHA took advantage of the massive amounts of human labor devoted to solving CAPTCHAs to read and digitize scanned archival texts.^[18][19]

About 200 million CAPTCHAs are solved by humans around the world every day. In each case, roughly ten seconds of human time are being spent. Individually, that's not a lot of time, but in aggregate these little puzzles consume more than 150,000 hours of work each day. What if we could make positive use of this human effort? reCAPTCHA does exactly that by channeling the effort spent solving CAPTCHAs online into "reading" books.^[19]

reCAPTCHA contributed to interpreting and digitizing the New York Times’ historical archive and, after its Google acquisition, assisted with Google Books efforts.^[19][20][21] The project team reports that reCAPTCHA-based text recognition is able to achieve accuracy of over 99.1% in correctly identifying individual words. Conversely, the optical character recognition software available at the time was only able to accurately identify 83.5% of words.^[18]

In precisely one year of reCAPTCHA deployment, human users collectively solved over 1.2 billion CAPTCHA tasks, translating over 440 million suspicious words that had been unreadable to optical character recognition software. (The same words were presented to multiple users, whose responses were then cross-verified before a word was confirmed.) Consequently, the authors ultimately characterized reCAPTCHA as a successful use of broad human computational power.^[18]

reCAPTCHA digitization efforts may be deemed as exploitative of human users who, in order to access sites and private accounts, are forced to dedicate their own unpaid time and brainpower—in however minuscule increments—to deciphering words for the archives of for-profit companies, even if the dissemination of that knowledge ultimately benefits them. One woman filed a lawsuit against Google on this basis, but the case was dismissed for the lack of damages, misrepresentation, and/or duty to disclose.^[22][23] Additionally, the prevalence and dominance of human computation and artificial intelligence digitization can contribute to the decline of professional human transcribers, whose jobs are rendered largely obsolete comparatively, as reCAPTCHA deployment can generate higher accuracy at lower price points.^[18][20][21]

Training Artificial Intelligence

In a similar vein, there has been speculation that image-based CAPTCHAs have been and are still used to tag image databases and train artificial intelligence systems in image recognition. This suspicion has been particularly heightened by Google's gradual shift towards street view images for its reCAPTCHA v2 tasks; rather than identifying fruit or animals, users are typically asked to identify crosswalks, traffic signals, cars and vans, or other street features.^[24][25][26] Such imagery could easily be used to train self-driving cars, which are a tech focus for Google and other companies. It's also consistent with standard machine learning feedback and training practices: humans label images for the computer to review, and the system decides for itself what features cause that image to fall under that categorization.^[24] This use of nonconsensual human computational trials has been viewed as coercively misleading on the part of tech companies.^[24][25] Evidence, however, is largely circumstantial, and Google has yet to confirm precisely how it uses reCAPTCHA v2 data.

Web Accessibility Concerns

CAPTCHA has been regularly criticized for impeding web accessibility and disadvantaging users with disabilities.^[10][11] Gaggi notes that users with disabilities may have particular incentive to utilize digital services.^[10] For instance, individuals with physical disabilities may have difficulties with transportation and mobility and so prefer to access services without leaving their homes, whereas individuals with auditory disabilities may prefer the text-based web and not wish to interact in-person where verbal communication will typically be the default. Consequently, it is of even more importance that such spaces remain accessible.^[10]

Disabilities

Todorov, Bogdanova, and Noev have created a list of common CAPTCHA approaches and the associated disabilities that could pose a conflict,^[11] with which Gaggi largely concurs:

• Character-Based: Distorted symbols can be difficult or impossible for people with visual disabilities or certain cognitive and learning disabilities (e.g., dyslexia) to interpret, and a disadvantage is posed for foreign language users or non-native speakers.^[10][11]
• Sound: Sounds of any kind, particularly distorted ones, can be difficult or impossible for Deaf people or those with auditory disabilities, such as auditory processing disorder, to interpret.^[10][11]
• Logic Puzzle: Depending on the nature of the puzzle, can be difficult for a wide assortment of individuals. For instance, implicature and meaning for autistic people or those with language disabilities, mathematical tasks for people with dyscalculia or low educational attainment, and virtually all sorts of puzzles for people with cognitive and intellectual disabilities generally.^[10][11]
• Biometrics and Cryptography: Can pose concerns for aging users, young users, and users with injuries or physical disabilities, for whom physical characteristics may not align with existing limitations of biometric software.^[11]
• Non-Interactive Methods: No serious problems are associated with non-interactive methods, but it should be noted that users with disabilities may engage in unusual behavior: for instance, favoring keyboards over mice or trackpads,^[10] filling out information at odd rates or in an odd order, and so forth.^[11]

Other Concerns

Myriad cultural barriers are presented, in addition to the disability barriers. Foreign language speakers or non-native speakers struggle with character- and audio-recognition tasks, and when a user is disadvantaged both culturally and disability-wise, they are even less likely to successfully solve the CAPTCHA.^[10][11] They might also struggle with puzzle tasks: for instance, someone from China may not be familiar with American traffic signals and devices or traditional Western breakfast foods. People with low educational opportunity or obtainment can struggle with a multitude of tasks, even in the absence of a clinically significant disability. Older adults and children may be confused by the request or purpose of the CAPTCHA.

Legality & Solutions

The United States government offers no specific interpretation or judgement of CAPTCHAs, but advises that websites are entirely subject to the Americans with Disabilities Act, and consequently, CAPTCHA accessibility constraints may violate federal law.^[27]

To counter CAPTCHA web accessibility concerns, researchers recommend that websites implement CAPTCHAs only when clearly indicated for security and integrity reasons, rather than by default on each site.^[11] They also recommend that there be no time limit within which to solve the CAPTCHA.^[11] Finally—and most importantly—sites should use passive CAPTCHAS where possible, that require no user interaction or input and run in the background.^[10][11] One study specifically advises the use of passive reCAPTCHA v3 (see below) risk assessment over reCAPTCHA v2, which employs active visual/auditory verification as a failsafe.^[10]

CAPTCHA Evasion

Artificial Intelligence Progression

Computers and programs have massively advanced over the past two decades, and have become increasingly capable of breaking and solving CAPTCHAs.^[8][9] Traditional text-based character-recognition CAPTCHAs are so reliably bypassed that they are now rendered functionally obsolete. ^[8] As such, CAPTCHAs are continually evolving, and developers and hackers are in a near-constant arms race—one that is expected to continue for the foreseeable future—to deploy new software and stay a step ahead.^[9]

In 2013, Google inadvertently cracked its own CAPTCHA service, reCAPTCHA, with a technology intended to improve Google Maps's street view. The algorithm can identify over 90% of street numbers on houses and buildings, and those skills are transferrable—it was able to identify text on Google's reCAPTCHA v1 tasks with 99.8% accuracy.^[28] The company unveiled a radically different model, reCAPTCHA v2, soon after.^[5] Unlike its predecessor, reCAPTCHA v2 involves image identification and categorization.^[5][6]

Substantial, although lesser, progress has also been made in the solving of newer image-based categorization CAPTCHAs. For instance, a group of researchers at Columbia University were able to achieve 70.78% of Google reCAPTCHA v2 image tasks.^[29]

Human Computation

Because most humans are reliably and easily able to bypass CAPTCHAs, mass-scale human labor can be used to bypass CAPTCHAs. The fruits of their labor are then handed immediately back to computer and bot accounts, in a delivery process that averages under twenty seconds. Middlemen recruit labor from low-income countries in Asia; in particular, Russia, India, China, Vietnam, and Bangladesh were indicated as probable sources.^[30] They are able to take advantage of a sizeable, low-cost workforce to sell one thousand solved CAPTCHAs for nominal, single-digit sums in United States dollars. This more or less defeats the purpose of CAPTCHAs and transforms anti-spam measures into a purely economic barrier. Furthermore, the wage and labor conditions posed to such workers present a potential ethical concern.^[30][31]

Timeline and Versions

The logo for Google's reCAPTCHA division and services.^[5] Image Credit: Google.

An example provided by Google of its reCAPTCHA v2 further confirmation task. The user is shown nine images depicting various foods and asked to select the images depicting oranges.^[6] Image Credit: Google.

While other programs exist and remain in use, reCAPTCHA iterations are fairly dominant over other models. One tech consulting firm estimates that Google reCAPTCHA captures 97.5% of the market share for CAPTCHAs.^[32] This is at least in part due to Google's own ubiquity and dominance across the internet—Google offers a host of design and security services for other companies,^[33] in addition to holding 90.8% of the market share for web searches.^[34] The company has been criticized for exercising this sort of monopolistic dominance, which poses pragmatic economic, social, and ethical concerns regarding inefficiency, access, and abuse of power;^[35] Google is currently facing an associated lawsuit from the United States Department of Justice.^[36]

CAPTCHA progression is not necessarily linear or mutually exclusive, and multiple editions remain in use simultaneously. For instance, Google continues to support three versions of reCAPTCHA.^[5]

reCAPTCHA

reCAPTCHA is a notable CAPTCHA service and company originally affiliated with Carnegie Mellon University's School of Computer Science.^[18][37] Luis von Ahn, who was a part of the original CAPTCHA team in 2000, is credited as the project’s executive producer.^[1][37] The project is the original team’s officially recommended CAPTCHA iteration.^[1]

von Ahn received a prestigious MacArthur Fellowship in 2006, primarily in recognition and support of his work with CAPTCHA.^[38][39] He stated that he planned to use the reward money to further his human computation ambitions.^[39] von Ahn ultimately accomplished this with his novel reCAPTCHA text-identification project (see Digitizing Information, above).^[18]

Google acquired the project in September 2009^[40] and has spearheaded successive versions (reCAPTCHA v2, reCAPTCHA v3, and reCAPTCHA Enterprise).^[5]

reCAPTCHA v1

reCAPTCHA v1 was a text-based implementation that tasked users with interpreting distorted and/or broken text.^[18][19] It has been used in the past to digitize text archives (see Digitizing Information, above); it is unclear whether it is still used for this purpose.^[18][19]

Google discontinued support for reCAPTCHA v1 in 2018, but continues to support and offer successive reCAPTCHAs.^[41]

reCAPTCHA v2

reCAPTCHA v2 is an image-recognition implementation. One of its features runs passively and requires no user input by default. Other options require active verification. If indicated, users are first prompted to click on and check a box, at which point some users will be verified as human and allowed to continue, while suspicious users will be presented with an additional task prior to verification.^[6][41] If a user is prompted to take an additional task, that task will be in the form of images a user must identify: for instance, by selecting all the images containing orange slices from a grid of nine food-related images.^[6][41]

reCAPTCHA v2 allows users to opt for an auditory task, rather than a visual task, and Google reports that it is compatible with major screen-reading services (see Web Accessibility Concerns, above).^[6] However, it remains unsupported for deaf-blind users.^[42]

References

1. ↑ ^{1.0 1.1 1.2 1.3 1.4} Carnegie Mellon University. (2000–2010). The Official CAPTCHA Site. http://www.captcha.net/
2. ↑ ^{2.0 2.1} Robinson, S. (2002, December 10). Human or Computer? Take This Test. The New York Times, F1. https://www.nytimes.com/2002/12/10/science/human-or-computer-take-this-test.html
3. ↑ ^{3.0 3.1 3.2 3.3} von Ahn, L., Blum, M., Hopper, N.J., Langford, J. (2003). CAPTCHA: Using Hard AI Problems for Security. In: Biham, E. (eds) Advances in Cryptology — EUROCRYPT 2003. EUROCRYPT 2003. Lecture Notes in Computer Science, vol 2656. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-39200-9_18
4. ↑ Wikipedia. Create Account. https://en.wikipedia.org/w/index.php?title=Special:CreateAccount
5. ↑ ^{5.0 5.1 5.2 5.3 5.4 5.5} Google. reCAPTCHA. https://www.google.com/recaptcha/about/
6. ↑ ^{6.0 6.1 6.2 6.3 6.4 6.5} Google. reCAPTCHA Help. https://support.google.com/recaptcha/?hl=en
7. ↑ Wikipedia. https://en.wikipedia.org/w/index.php?title=Special:CreateAccount
8. ↑ ^{8.0 8.1 8.2 8.3} Gao, Y., Gao, H., Luo, S., Zi, Y., Mao, W., Wang, P., Shen, Y., & Yan, J. (2021, August 11–13). Research on the Security of Visual Reasoning CAPTCHA. Proceedings of the 30th USENIX Security Symposium. 978-1-939133-24-3. https://www.usenix.org/conference/usenixsecurity21/presentation/gao
9. ↑ ^{9.0 9.1 9.2 9.3} Zhang, Y., Gao, H., Pei, G., Luo, S., Chang, G., & Cheng, N. (2019, August 5–8). A Survey of Research on CAPTCHA Designing and Breaking Techniques. Proceedings of the 2019 18th IEEE International Conference On Trust, Security And Privacy In Computing And Communications/13th IEEE International Conference On Big Data Science And Engineering (TrustCom/BigDataSE), Rotorua, New Zealand. https://ieeexplore.ieee.org/document/8887344
10. ↑ ^{10.00 10.01 10.02 10.03 10.04 10.05 10.06 10.07 10.08 10.09 10.10} Gaggi, O. (2022, June 28). A study on Accessibility of Google ReCAPTCHA Systems. Oasis 2022. Conference Proceedings. Barcelona, Spain. Association for Computing Machinery. https://dl.acm.org/doi/pdf/10.1145/3524010.3539498
11. ↑ ^{11.00 11.01 11.02 11.03 11.04 11.05 11.06 11.07 11.08 11.09 11.10 11.11} Todorov, T., Bogdanova, G., & Noev, N. (2022, July 25). Accessibility Challenges with CAPTCHA Services in Digital Libraries. Digital Presentation and Preservation of Cultural and Scientific Heritage. Conference Proceedings. Vol. 12. Sofia, Bulgaria. Institute of Mathematics and Informatics.
12. ↑ ^{12.0 12.1 12.2 12.3 12.4} Turing, A.M. (1950, October). Computing Machinery and Intelligence, Mind, 59(236), 433–460. https://phil415.pbworks.com/f/TuringComputing.pdf
13. ↑ ^{13.0 13.1 13.2} Metz, C. (2023, January 25). How Smart Are the Robots Getting? The New York Times, https://www.nytimes.com/2023/01/20/technology/chatbots-turing-test.html
14. ↑ ^{14.0 14.1 14.2 14.3 14.4} Eliot, L. (2020, July 20). The Famous AI Turing Test Put In Reverse And Upside-Down, Plus Implications For Self-Driving Cars. Forbes. https://www.forbes.com/sites/lanceeliot/2020/07/20/the-famous-ai-turing-test-put-in-reverse-and-upside-down-plus-implications-for-self-driving-cars
15. ↑ Gibbs, S. (2014, October 27). Elon Musk: artificial intelligence is our biggest existential threat. The Guardian. https://www.theguardian.com/technology/2014/oct/27/elon-musk-artificial-intelligence-ai-biggest-existential-threat
16. ↑ Clark, S. (2014, December 2). Artificial intelligence could spell end of human race – Stephen Hawking. The Guardian. https://www.theguardian.com/science/2014/dec/02/stephen-hawking-intel-communication-system-astrophysicist-software-predictive-text-type
17. ↑ Dredge, S. (2015, January 29). Artificial intelligence will become strong enough to be a concern, says Bill Gates. The Guardian. https://www.theguardian.com/technology/2015/jan/29/artificial-intelligence-strong-concern-bill-gates
18. ↑ ^{18.0 18.1 18.2 18.3 18.4 18.5 18.6 18.7} von Ahn, L., Maurer, B., McMillen, C., Abraham, D., & Blum, M. (2008, September 12). reCAPTCHA: Human-Based Character Recognition via Web Security Measures. Science, 321, 1465–1468. 10.1126/science.1160379
19. ↑ ^{19.0 19.1 19.2 19.3 19.4} reCAPTCHA. (2009). What Is reCAPTCHA. https://web.archive.org/web/20100611210239/http://recaptcha.net/learnmore.html
20. ↑ ^{20.0 20.1} Gugliotta, G. (2011, March 28–29). Deciphering Old Texts, One Woozy, Curvy Word at a Time. The New York Times, D3. https://www.nytimes.com/2011/03/29/science/29recaptcha.html
21. ↑ ^{21.0 21.1} Stone, B. (2009, September 16). Google Buys Service That Uses Humans to Digitize Books. The New York Times. https://archive.nytimes.com/bits.blogs.nytimes.com/2009/09/16/google-buys-service-that-uses-humans-to-digitize-books/
22. ↑ Harris, D. (2016, January 2015). Massachusetts woman's lawsuit accuses Google of using free labor to transcribe books, newspapers. Boston Business Journal. https://www.bizjournals.com/boston/blog/techflash/2015/01/massachusetts-womans-lawsuit-accuses-google-of.html
23. ↑ Dinzeo, M. (2016, February 5). Google Ducks Gmail Captcha Class Action, Courthouse News Service, https://www.courthousenews.com/google-ducks-gmail-captcha-class-action/
24. ↑ ^{24.0 24.1 24.2} Maruzani, R. (2021, January 26). Are You Unwittingly Helping to Train Google’s AI Models? Towards Data Science, Medium. https://towardsdatascience.com/are-you-unwittingly-helping-to-train-googles-ai-models-f318dea53aee
25. ↑ ^{25.0 25.1} Healey, M., & Flores, M. CAPTCHA If You Can. Ceros. https://www.ceros.com/inspire/originals/recaptcha-waymo-future-of-self-driving-cars/
26. ↑ Vega, E. (2021, May 14). Why captchas are getting harder [Video]. Vox. https://www.vox.com/22436832/captchas-getting-harder-ai-artificial-intelligence
27. ↑ ADA.gov. (2022, March 18). Guidance on Web Accessibility and the ADA. U.S. Department of Justice Civil Rights Division. https://www.ada.gov/resources/web-guidance/
28. ↑ Goodfellow, I. J., Bulatov, Y., Ibarz, J., Arnoud, S., Shet, V. (2014, April 14). Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks. arXiv. https://doi.org/10.48550/arXiv.1312.6082
29. ↑ Sivakorn, S., Polakis, I., & Keromytis, A.D. (2016, March 21–24). I am Robot: (Deep) Learning to Break Semantic Image CAPTCHAs. Proceedings of the 2016 IEEE European Symposium on Security and Privacy (EuroS&P), Saarbruecken, Germany. http://www.cs.columbia.edu/~polakis/papers/sivakorn_eurosp16.pdf
30. ↑ ^{30.0 30.1} Motoyama, M., Levchenko, K., Kanich, C., McCoy, D., Voelker, G., & Savage, S. (2010). Re: CAPTCHAs – Understanding CAPTCHA-Solving Services in an Economic Context. Proceedings of the 19th USENIX Security Symposium, Washington, DC, USA. https://klevchen.ece.illinois.edu/pubs/mlkmvs-usesec10.pdf
31. ↑ International Labour Organization. "World Employment and Social Outlook: Trends 2019." (2019, February 13). United Nations. https://www.ilo.org/wcmsp5/groups/public/---dgreports/---dcomm/---publ/documents/publication/wcms_670542.pdf
32. ↑ Slintel. (2023, February 1). reCAPTCHA. https://www.slintel.com/tech/captcha/recaptcha-market-share?
33. ↑ Google. Google Cloud. https://cloud.google.com/
34. ↑ Desjardins, J. (2018, April 23). How Google retains more than 90% of market share. Business Insider, https://www.businessinsider.com/how-google-retains-more-than-90-of-market-share-2018-4
35. ↑ Vaitilingam, R. (2020, November 24). Does Google Have Too Much Market Power? Chicago Booth Review, University of Chicago. https://www.chicagobooth.edu/review/does-google-have-too-much-market-power
36. ↑ U.S. Department of Justice. (2023, January 24). Justice Department Sues Google for Monopolizing Digital Advertising Technologies. https://www.justice.gov/opa/pr/justice-department-sues-google-monopolizing-digital-advertising-technologies
37. ↑ ^{37.0 37.1} reCAPTCHA. (2009). About Us. https://web.archive.org/web/20100611210259/http://recaptcha.net/aboutus.html
38. ↑ MacArthur Foundation. (2006, September 1). Luis von Ahn. https://www.macfound.org/fellows/class-of-2006/luis-von-ahn#searchresults
39. ↑ ^{39.0 39.1} Spice, B. (2006, September 18). Brilliant Young Scientist Luis von Ahn Earns $500,000 MacArthur Foundation "Genius Grant”. Carnegie Mellon Today, Pittsburgh, PA, https://www.cmu.edu/cmnews/extra/060918_ahn.html
40. ↑ von Ahn, L., & Cathcart, W. (2009, September 16). Teaching computers to read: Google acquires reCAPTCHA. Google. https://googleblog.blogspot.com/2009/09/teaching-computers-to-read-google.html
41. ↑ ^{41.0 41.1 41.2} Google. (2021, June 1). Choosing the type of reCAPTCHA. https://developers.google.com/recaptcha/docs/versions
42. ↑ Google. What is CAPTCHA? https://support.google.com/a/answer/1217728?hl=en