Cloudflare said on August 31, 2026 that it launched Adaptive Intelligence, a new bot detection engine inside Cloudflare Bot Management. For teams already using Cloudflare bot management, the component shipping first replaces the fixed, periodically released machine learning model behind bot score with one Cloudflare says retrains continuously on live traffic — which changes how much manual tuning login-abuse and scraping defenses may need.
What changed
Cloudflare’s August 31 blog post announces Adaptive Intelligence as a new bot detection engine; SiliconANGLE and Investing.com report that it is integrated into Bot Management.
The post describes three components for the engine and says only the first is launching now: the machine learning at the center of bot score, retraining continuously instead of shipping as a fixed version. The other two — automatic generation of short-lived “disposable” rules deployed and retired at random intervals, and learning from scoring corrections across customer traffic — are described in the same post as “soon to follow.” Secondary coverage did not draw that line, describing the engine as one that already writes, rotates, and discards its own blocking rules mid-attack; in the primary source that is design intent for a component still to come.
Cloudflare also says Adaptive Intelligence sits behind bot score. In the company’s description, bot score already combines machine learning, behavioral validation, JavaScript fingerprinting, heuristics, and verified-bot checks.
Cloudflare says it analyzes more than a trillion requests a day for signs of automated abuse, and that the engine retrains on that live traffic.
Why B2B teams should care
Cloudflare’s operational argument is that attackers adapt continuously while defenses ship in discrete releases. SiliconANGLE reported Cloudflare’s framing that conventional tools can depend on scheduled updates taking weeks or months, a mismatch when bot operators change tactics faster than defenders refresh rule sets.
That matters most where attack economics already favor the attacker. SiliconANGLE reported Dane Knecht’s view that fixed defenses fail when attack scaling costs are effectively zero, while Cloudflare’s blog says Adaptive Intelligence is designed to make attacks too slow and costly to continue. Investing.com separately reported that Knecht described the product as creating a moving target.
The attack categories named across the coverage are credential stuffing and scraping. Cloudflare says the engine evaluates traffic over several time windows at once — a short one to catch a developing burst, a longer one to tie scattered requests from thousands of addresses back to a single source — which is how it says it surfaces slow campaigns that stay under conventional rate thresholds.
Who is affected
SiliconANGLE reported that Adaptive Intelligence is available now to Bot Management customers. Cloudflare’s own post is more specific about the mechanics and narrower about the tier: it tells Enterprise customers to turn on “Auto Update Machine Learning” in the Bot Management dashboard, and says that with the setting on there is no model version to migrate, nothing to configure, and the bot score teams already build on keeps working unchanged. SiliconANGLE said Cloudflare did not disclose pricing; Investing.com reported that the company disclosed neither pricing details nor specific availability dates.
As an editorial read grounded in the attack types named, the feature matters most to SaaS platforms, ecommerce properties, and marketplaces exposed to login abuse, scraping, and automated fraud.
What teams should check now
The first thing to check is the setting itself. Cloudflare says enablement runs through “Auto Update Machine Learning” in the Bot Management dashboard, so teams that left auto-updating off, or inherited a configuration someone else set, will not pick up the new engine by default.
Beyond that, teams should review how much they currently depend on static custom bot rules, manual tuning cycles, and bot score thresholds. Because Adaptive Intelligence sits behind bot score, any bot score-based policy — login protection, challenge decisions, automated traffic filtering — is where a change in the underlying engine would show up first.
Cloudflare’s answer to the obvious objection, that a model updating itself could start scoring real visitors differently, is validation: it says each new version first runs in shadow mode alongside the current one, scoring live traffic without affecting visitors, and that a version that would score real people worse does not go live. That is a claim about Cloudflare’s own release process, not a metric customers can audit.
Teams that enable or rely on the new engine should watch:
- false positives
- challenge and block actions
- attack-specific outcomes on login flows
- attack-specific outcomes on content endpoints
What remains unclear
- Not yet confirmed: pricing for Adaptive Intelligence. SiliconANGLE reported that Cloudflare did not disclose it.
- Not yet confirmed: when the second and third components — disposable rule generation and learning from scoring corrections — actually ship. Cloudflare says only that they are “soon to follow.”
- Not yet confirmed: whether the engine reaches Bot Management customers below the Enterprise plan, the only tier named in Cloudflare’s enablement instructions. SiliconANGLE reported availability to Bot Management customers now; Investing.com reported no specific availability dates.
- Not yet confirmed: geographic rollout limits or whether availability differs by region or account scope.
- Not yet confirmed: quantified efficacy metrics such as false-positive reduction, detection-rate improvement, or the amount of bot traffic blocked.
What to watch next
SiliconANGLE reported that Cloudflare said Precursor became generally available in July, and Investing.com reported that Adaptive Intelligence combines Precursor session signals with network telemetry. Cloudflare’s post presents the two as halves of one design: Precursor measures how a visitor behaves in the browser across a session, while Adaptive Intelligence learns from detection signals across the network. The question worth tracking is whether Cloudflare is assembling a broader bot and session-risk stack rather than isolated rule packs — a direction other security vendors are taking too, as with Cato Networks and its agentic threat prevention launch.
The concrete milestone is the second component. Cloudflare’s case — against a bot share SiliconANGLE reported at roughly 57% of web requests in July — rests on rules that appear and vanish faster than an attacker can map them. Until disposable rule generation ships, Bot Management customers have a detection model that updates itself, not a rule set that rewrites itself during an attack.
