Cloudflare’s Adaptive Intelligence Turns Bot Detection Rules Into a Moving Target

Cloudflare on August 31 introduced Adaptive Intelligence, a machine learning detection engine built into its Bot Management product that retrains itself on live traffic and writes short-lived, disposable rules instead of the static, hand-maintained rules bot defense has relied on for years. The pitch, laid out by Cloudflare CTO Dane Knecht, is blunt: “Building taller walls fails when the cost of scaling an attack is effectively zero.” Adaptive Intelligence, he said, “creates a moving target, rendering an attacker’s engineering efforts obsolete.”

Why Static Rules Lose

Cloudflare’s own explanation of the problem is a fair summary of how bot defense has worked industry-wide: write a rule to catch a technique, wait for the attacker to study the signal and route around it, then write another rule. “Bot detection has always answered a new attack technique by writing a rule to catch it,” the company’s engineering blog post explains. “That works, until the attacker studies the signal, learns how to circumvent it, and forces another rule to be written.”

The deeper flaw, according to Cloudflare, is that deterministic systems are predictable by design: a rule that always produces the same output for the same input hands an attacker a stable target to probe and defeat. In the company’s words, “a defense that never changes teaches the attacker how to beat it.” That predictability problem has gotten worse as commodity AI tools make it cheap for low-skill attackers to generate new evasion techniques, and as bot operators increasingly trade tactics with each other or simply buy access to commercial evasion services.

Three Ways the Engine Adapts

It retrains continuously

Instead of shipping model updates on a release schedule, Adaptive Intelligence retrains on live traffic as it arrives. Cloudflare frames the payoff in immediacy: “A technique that shows up this week is one the engine can recognize this week,” rather than waiting for the next scheduled model push.

It ships disposable rules

Where a traditional bot rule stays in production indefinitely once it is written, Adaptive Intelligence is built to churn. Cloudflare says the system is “designed to create disposable rules aimed at a specific attack, deploy and retire them at random intervals, and never leave them in place long enough to become a fixed target.” A rule that exists for hours instead of months gives an attacker far less time to reverse-engineer and evade it.

It learns from customer corrections

The engine also treats human feedback as training data. When a customer flags a legitimate visitor that Adaptive Intelligence scored incorrectly, Cloudflare says that correction feeds back into the model as a training signal, tuning detection toward the specific traffic patterns a given customer actually sees rather than a one-size-fits-all baseline.

Testing Changes Before They Touch Real Traffic

Retraining a live detection model on live traffic is a risky operation if a bad update starts blocking real visitors, so Cloudflare built a staged rollout around it. New model versions run in what the company calls shadow mode first: “Before a new version becomes your primary defense, it runs in shadow mode alongside the current one, scoring live traffic without affecting a single visitor.” Only after that validation period does a new version graduate to production, and even then Cloudflare says it watches score distributions, challenge outcomes, and customer feedback so a bad rollout can be paused or rolled back before it reaches a customer’s whole network.

A Second Layer: Precursor

Adaptive Intelligence does not stand alone. Cloudflare launched a companion product called Precursor in July, a client-side engine that runs inside the browser to continuously monitor an entire visitor session, tracking signals like mouse movement, scrolling rhythm, typing cadence, clipboard activity, and page visibility duration rather than checking identity once at the door. Knecht made the same case for Precursor at its launch, telling Help Net Security: “Traditional security checks look at a single moment in time, but modern bots have gotten smart enough to fake their way through the front door.” Instead, he said, “we are looking at behavior over the entire visit.”

The two products are designed to reinforce each other. Cloudflare describes the split this way: “Precursor does so through measuring continuous session-behavior; Adaptive Intelligence learns from bot detection signals across the whole network, and the signals from one make the other harder to fool.” In practice, that means a bot sophisticated enough to fake session behavior in the browser still has to get past network-wide pattern detection, and vice versa.

The Economics Behind the Announcement

Cloudflare says it now analyzes more than a trillion requests a day for signs of automated abuse, the scale that makes continuous retraining possible in the first place. When Cloudflare launched Precursor in July, several outlets covering that announcement cited a widely repeated figure: automated bot traffic had already grown to roughly 57 percent of web requests, more than the share coming from humans. Neither Cloudflare’s Precursor post nor its Adaptive Intelligence post states that number directly or names an underlying dataset, so it is worth treating as an industry estimate that has circulated in coverage rather than a figure Cloudflare has formally published and sourced.

Independent research cited by PYMNTS gives a narrower, better-sourced picture of who is affected most. According to PYMNTS Intelligence research, 60.6 percent of financial services firms reported increased bot traffic over the past year, the highest rate of any industry PYMNTS tracked. The same research estimated that bot-driven gaps in automated identity verification cost businesses roughly $95 billion in lost revenue globally each year, an average of 3.1 percent of annual revenue per affected enterprise. PYMNTS also flagged what it called a confidence paradox: 96.3 percent of security executives said they were confident in their bot defenses, while 90 percent simultaneously reported real operational problems caused by hostile automated traffic.

What’s Available Now, and What’s Next

Adaptive Intelligence is currently scoped to Cloudflare’s Enterprise customers. Turning it on does not require picking a model version or scheduling a migration: customers enable a setting called “Auto Update Machine Learning” in the Bot Management dashboard, and Cloudflare says “with it on, you get Adaptive Intelligence automatically, with no version to migrate and nothing to configure.”

Cloudflare frames this launch as a starting point rather than a finished product, saying “more of the engine comes online from here.” The company lists three areas of near-term work: expanding how many detections the engine generates automatically, connecting more data points from across its network into the model, and adding more response options beyond the block-or-allow decisions bot management systems have traditionally offered. For a company that already sits in front of a large share of the web’s traffic, the more consequential claim in this launch may not be the technology itself but the economic argument underneath it: that constantly moving the target costs Cloudflare less compute than it costs an attacker to keep rebuilding tooling from scratch.