Continuing on StyloBot this weekend, staging site showed a regression after my new work on Browser Modes so working through that - code llm drifted away from a core pattern in a big refactor...as it always does.

Core lesson: make your test apps *fail loudly* LLMs are great at papering over spreading, structural cracks in your system.

Distillation phase; not planned work as it's responding to live detection issues. Where a single homan controlled browser sends different headers, request 'shape' for stuff like data / resources / api calls / content.
Each of those is a different 'mode' which NORMALLY bot detectors will just toss the strange differences and merge into a single 'identity based' fingerprint.

*However* in StyloBot it can tell them apart because it doesn't throw away these extra discriminators. The bug appeared like multiple clients requesting from the same IP but different 'deceptive' browsers (as I expected all the headers etc). Same goes for AdBlock on/off...it changes what your browser *looks like* by stripping headers etc...
So instead I actually detect these 'modes' and use them to add to discriminaton signal; the proportion and timing cadence is SUPER hard to fake for automation; humans run browsers in 'noisy' computational environments, bots tend to be synthetically 'cleaner'.

Anyway that's my weekend, the code llm decided (as it tends to) to layer another parallel system on top of the existing one and it SUCKS - 2s a response as it does a full *db* match against ALL our 'prototyes' each request.
It got the idea but forgot the pattern; I use a single consolidated behaviourally shaped LFU 'transparent' - you don't worry about DB just read and wwrite through the cache and it handles updates etc - LAZY CQRS with none of the guarantees for *super* high speed & in memory.
As usual this is an *odd* archutecture of this sub-feature so the Code LLM (Calude) SMEARS it towards the normative for this.
I told it new in memory stores are not allowed; as it will ALWAYS add these (IMemoryCache) in ASP.NET apps...try and stop it!
So it decided this was a direct to DB operation; it followed one constraint then SKIPPED all the others.
#stylobot #botdetection #botprotection

👀 What's being cooked at CrowdSec?

Your WAF already knows *what* requests are doing.

What if it could also help answer *who* is behind them?

More soon!

#CyberSecurity #WAF #BotDetection #ThreatIntelligence

StyloBot today; more work on Endpoint UX - for ASP.NET core pack you get the ability to live update endpoint policies. So working out the best UX.
Almso more on the new 'browser mode' work; need to write a blog post on that SEEMS to be anew technique, using the proportion of browser mode changes combined with markov paths to detect subtle timing errors automation lacks.
Even stuff like how a users' browser runs on a desktop OS in terms of xhr, resource requests etc is influenced by other applications loaded, gpu presence etc.
NOT a 'fingerprint' per-se (though it could maliciously be used that way) but a parretn StyloBot can connect to human and bot signatures to find the divergence between the two.
Useful as it's also per-site and likely load dependent so it's a nicely patternable signature - NOT repeatable so not 'this is definietely Scott on his Chrome Mac', that's *identity* no, behaviour; so it *varies* in a human way.
And because StyloBot already watches centroid drift generally, browser modes become another way to detect subtle mismatches: not "is this request bad?" but "does this client’s behavioural movement make sense for the browser it claims to be?"

#stylobot #botdetection #zeropii

University of Missouri: User-friendly bot detection. “Traditional bot detection methods like CAPTCHA are effective, but they come at the cost of user experience. Our team wanted to explore whether machine learning could handle the same job entirely in the background — no interruptions, no puzzles, just seamless verification. Beyond bot detection, we saw potential for this approach to extend […]

StyloBot free day as I ran myself ragged trying to get it going in my free time (very little of which I HAD finishing up 2x contracts!).

Biggest win is dropping the ONNX dependency.

Earlier versions used ONNX embeddings as a shortcut: turn a client signature into a vector and compare it.

It worked, but it was never quite the right abstraction. Embeddings are built for language. StyloBot’s inputs are behavioural structures.

The new version defines that behavioural vector space directly. Requests, sessions, browsers, bots, scrapers, and odd clients are placed into a real StyloBot-native space. The system ships with archetype centroids, then adapts those centroids to the actual traffic it sees.

So instead of asking a model what a client 'means', StyloBot learns what your traffic looks like.

StyloBot is REALLY a conceptually unfolded ML model so it sort of trains itself on real traffic around centroids and updates as it goes. It's ODD.

Now out in Release Candidate https://github.com/scottgal/stylobot/releases

Plan is still for full release June 1st but the FOSS client MAY reach RTM quality before that (lots of manual testing!)

#BotDetection #CyberSecurity #DotNet #SQLiteVec #VectorSearch #BehaviouralInference #AIInfrastructure #OpenSource

New article about StyloBot - my FREE realtime adaptive automation blocker and detector.

https://www.mostlylucid.net/blog/stylobot-fingerprint

Trying to put together a *human* level descriptor as I prepare to make it my full time gig (amongst other self-released projects).

#botdetection #fraud #stylobot

Getting stuck in an infinite “please try again” loop while creating a Microsoft account.
It surely does a decent job blocking batch registrations too, right?

#Microsoft #AccountSecurity #SpamPrevention #BotDetection #Tech

Clean analytics start with blocking fake traffic.

→ Real-time bot detection & blocking
→ Live analytics dashboard

Get started free → gobot.cx

#Gobot #BotDetection #CleanTraffic

One line of code. Real-time bot protection.

→ Real-time bot detection & blocking
→ 3 firewall levels: Low / Medium / Hard

Protect your website → gobot.cx

#Gobot #BotDetection #CleanTraffic

Real data from a Gobot user:

✅ 28,233 real visitors
🚫 73,938 bots blocked
📊 102,171 total

72.4% of traffic was bots — all caught automatically.

Try it → gobot.cx

#Gobot #BotDetection #WebSecurity