Show HN: Roundtable (YC S23) – Survey Fraud and Bot Detection API
Hi HN, we're Mayank and Matt of Roundtable (www.roundtable.ai). We launched our Survey Simulator as part of S23 (see Launch HN: https://news.ycombinator.com/item?id=36865625), and ultimately decided to pivot because we couldn't convince users to trust our results and our value proposition wasn't clear.
We conducted tons of user discovery interviews, and a lot of big players talked about the pain of manual QC and bot detection. We launched a survey fraud and bot detection API (Alias; docs.roundtable.ai) that determines whether a bot is answering your survey or not. Our API embeds Javascript into survey programs (Forsta, Qualtrics, jsPsych, etc.) to track user behavior and make a judgment on whether the data is from a high-quality participant.
You can use our demo (and try to break it) here: www.roundtable.ai/hn-demo! See if you can generate LLM-based text without any flags going off
(
9 comments
| 3 points
| on HN
)
Do any tools exist to detect suspicious behaviour for e.g. multiple choice surveys? I'm thinking of tab switching to Google for an answer for instance.
First, your payload is being sent using base64 encoded using the default alphabet. You only need a simple atob() to decrypt your payload.
Second of all, your bot detection script is very readable making the job of the attacker relatively easy to reverse.
Third, but not least, you do not have enough signals/fingerprints which means that your false positives are going to sky-rocket.
Bots are not dumb, they are programmed by real humans, your site is extremely easy to reverse. You need to add more obfuscation, more signals, better client-side protection in order to quality for real "bot detection".
Source: I reverse antibots for fun and profit, is literally all I've been doing for the past 2 years straight.
the arms race between bot and anti-bot is fascinating and I think I could reasonably overcome barriers like it, so HMU!