Why spam still gets past your CAPTCHA: solving services and AI

Why your CAPTCHA lets spam through

A CAPTCHA only asks: "can whoever is sending this solve the puzzle?". For a spammer, that is a technical problem with two well-known solutions: software that solves the puzzle, or a person who solves it for them. Both are sold as services.

Human solving services

A spam bot fills in your form, and when it meets the CAPTCHA it forwards the puzzle to a solving service. A worker somewhere sees the image, answers it within seconds, and the bot submits your form with the correct answer. To your website it looks exactly like a person solved it, because a person did.

Researchers at UC San Diego studied this market in detail and found solutions sold at retail for around one dollar per thousand, with the workers earning a small fraction of that [1]. In their tests, most services solved the CAPTCHAs of popular websites correctly more than 70% of the time, within about 20 seconds [2]. At those prices, sending thousands of spam messages through a CAPTCHA costs a spammer a few dollars.

Software solvers and AI

For many CAPTCHA types, spammers no longer need people at all. The 2023 study of modern CAPTCHAs compared human results with the best published attacks [3]:

• Distorted text: machines reach 99.8% accuracy in under a second; people managed 50–84%.
• reCAPTCHA checkbox: 100% accuracy in about 1.4 seconds, using automatically created "trusted" cookies [4].
• hCaptcha images: 98% for an attack, against 71–81% for people.
• Audio challenges: defeated with ordinary speech recognition [5].

Even Google's invisible reCAPTCHA v3, which scores visitors instead of showing puzzles, has been gamed: researchers reported an attack that earned high "human" scores 97% of the time [5].

Why a harder CAPTCHA doesn’t help

The natural reaction is to turn the CAPTCHA up to its hardest setting. The research shows why that backfires: on hCaptcha's difficult setting, human accuracy fell from 81.4% to 70.6% [3], while attackers simply adapt. A harder puzzle mostly moves the cost from the spammer to your customers. What people who give up cost you is covered in is CAPTCHA costing you customers?

What actually stops it

The answer is checks that don't depend on anyone solving anything, run on your own server where spammers can't see them:

1. A signed, one-time token issued when a real visitor starts the form. Bots that post straight to your form's address don't have one, and a copied token works only once.
2. A minimum time between opening the form and sending it.
3. A hidden trap field that automated form-fillers fill in.
4. Link rules, because most spam exists to place links.
5. A rate limit per visitor, which stops floods, including from paid human spammers.
6. A spam log, so you can see what was blocked and why.

Nothing automatic can stop a real person who is paid to paste an advert into your form. But rate limits and link rules make it slow and unprofitable, and none of it costs your customers anything. See how to stop bots from submitting your contact form.