Kernel

EngineeringCatherine Jue+Ulzii

Bot detection is dead, long live verified identity

Learn why, as agents blur the line between humans and automation and the economics of reverse‑engineering bot detection shift, verified identity is becoming the future.

Bot detection is dead, long live verified identity

bot detection has been a central part of our industry. it’s how the internet policed itself. it decided which signups were real, which login attempts were worth a second look, which ad impressions were safe to buy, and which purchases to approve. underneath all the vendors and acronyms was a simple assumption: humans on the internet are good; automation—associated with inventory hoarding, fake accounts, scraping, and fraud—is bad. agents broke that assumption. once you have agents that can go off and do work on your behalf, the line between “user” and “bot” blurs.

on top of that, today’s bot detection systems are built for a different adversary. vendors have spent years building increasingly sophisticated client-side fingerprinting: bytecode vms, obfuscated string tables, encrypted telemetry, crash-and-capture integrity probes. the underlying bet was straightforward: complexity buys time. deobfuscation is slow, tedious work, so by the time someone reverse-engineers your sensor script, you have already shipped a new one. agents have changed the economics of reverse-engineering. we saw this firsthand while debugging why a widely deployed bot-detection vendor was intermittently rejecting sessions from an otherwise clean browser of ours. in a matter of hours, one of our security engineers, armed with an agent, understood what its sensor script checked.

agents broke both halves of the old world: the overly simplistic assumption that "humans good, automation bad," and the comfortable lie that "complexity buys you time." what comes next won't look like a binary filter in front of your site, it will look more like verified identities for a mixed population of humans and agents, where actors can prove who they are.

to make this argument less theoretical, we’re sharing, at a high level, how quickly an agent-assisted engineer can understand a modern sensor script. at every step, agents took what used to be days of manual work and compressed it into hours.

what agents changed

modern sensor scripts are built to be expensive to read: custom bytecode vms, encoded string tables, encrypted telemetry. a year ago, making sense of one meant days of careful work from a researcher who had seen enough obfuscation to recognize the patterns. with an agent, it took minutes: the agent recognized the encoding scheme, wrote a decoder, and summarized what the script checks.

the same is true further down the stack. browser engines are millions of lines of c++, and agents make a codebase that size navigable in an afternoon instead of a week.

verified identities > bot detection

if a small team with off‑the‑shelf agents can deobfuscate your client‑side javascript in an afternoon, the security value of obfuscation is already trending toward zero. the future of defending the internet is not better detection tricks. it is verified identity.

that shift is better for everyone. website operators get reliable signals about who is accessing their services instead of probabilistic guesses. agent developers get a legitimate path that does not involve reverse-engineering sensor scripts. and users stop getting caught in the crossfire of false positives from increasingly aggressive bot detection.

that’s the path we’re building on. kernel identifies its agents with web bot auth, an emerging ietf standard that lets automated clients cryptographically sign their requests, so a site can verify who is behind the traffic and decide how to treat it. kernel is listed in cloudflare’s and vercel’s bot directories, and most major cdns, wafs, and bot-management services verify web bot auth signatures.

if you want to collaborate on the future of agent identity, reach out to partnerships@kernel.sh

more blog posts

view all