Your scanner just flagged 4,000 new vulnerabilities, 78 of them critical. Which one do you fix first?

To answer that question, Cloudflare is announcing early access to Vulnerability Discovery and Remediation, now part of Cloudflare Managed Defense. Vulnerability Discovery and Remediation is a new, invitation-only Cloudflare service that helps customers detect and mitigate vulnerabilities in their codebases.

Through the OpenAI Daybreak Defense Network, we use OpenAI Daybreak models, including GPT-5.6 Cyber, for reconnaissance, hunting, and validation against codebases that you authorize us to access. If we detect a vulnerability, we will then propose solutions to you, automatically checking each proposed patch and any accompanying proposed mitigation before presenting them for review. Importantly, you are in the driver’s seat: while we may propose code patches and other mitigations, you decide whether they are implemented.

Choosing what to fix first has always been hard. It's getting harder. Large language models can now surface weaknesses across a codebase in minutes, which means the number of findings keeps climbing. But the real problem is speed. Attackers can use AI to accelerate parts of vulnerability discovery and exploitation, giving security teams and developers less time to decide what matters and act on it.

Imagine that your scanner tells you there's a vulnerability in a handler. It doesn't tell you whether that code is deployed. It doesn't tell you whether anyone is actually hitting that route, what security activity surrounds it, or what controls you already have in place. You have to prioritize the finding without evidence of its production exposure or the protections already in place.

This is where we can help. With our global network, we can see which routes are active, how much traffic they carry, and what security events surround them. When customers enable Vulnerability Discovery and Remediation with Web Application Firewall (WAF), we can also see what rules are already applied and are actively blocking attacks. That context turns a generic finding into a specific priority: this vulnerability is in code that's live, on a route that's heavily used, with recent attack activity and no existing protection. And we can help you mitigate that vulnerability by proposing custom WAF mitigations and code patches tailored to your systems.

If this sounds familiar, it should. In “Build your own vulnerability harness”, we described the model-agnostic pipeline we use to scan Cloudflare's fleet, adversarially validate every finding, and turn raw model output into fixes engineers can trust. That internal system is one pillar of Vulnerability Discovery and Remediation. The harness gave us a way to find bugs at fleet scale. Vulnerability Discovery and Remediation brings that discovery process to the code the customer authorizes us to inspect, then connects the findings to production traffic, security events, and the edge controls that can act on them.

This diagram provides an overview of our process, which we explain in more detail below.

Adding context to a vulnerability harness

Our solution works across Cloudflare Workers and proxied applications. The process of detecting vulnerabilities begins with the collection of a traffic and security data snapshot from Web Assets and WAF. The snapshot shows which routes are active, how much traffic they receive, and whether recent security events are associated with them. For instance, a path exhibiting a high volume of detection triggers may also be considered critical for security context purposes. Web Assets and WAF itself serve as the first and second pillar of Vulnerability Discovery and Remediation respectively.

Next, we use source code vulnerability analysis to identify potential weaknesses in code. But that analysis does not show which routes reach it, how much traffic those routes receive, whether they receive suspicious requests, or which protections already apply. We treat routes carrying a high volume of requests as hot paths. Source code deployed to these routes undergoes stricter security profiling. Together, these signals provide evidence about how the API is used and where a vulnerability may be exposed.

For Workers, we retrieve the most recent source version of the Worker and its configured routes to identify the endpoints the Worker serves. Next, we match the Worker's routes to Web Assets and request metadata from Workers Observability, tying the exact source under review to the endpoints it handles in production. This collected network context stays available throughout the investigation, allowing agents to pull it when they need it. 

Our vulnerability harness then starts up. It begins by using the Reconnaissance agent to map request paths to the parts of the codebase that handle them. Reconnaissance uses that map to send hunter agents into specific sections of the customer-authorized code, where they look for vulnerabilities and pull in relevant network context as needed. That context can help the hunter agents pay more attention to code behind an active or recently targeted route, but it does not establish that a vulnerability exists. Every vulnerability finding has to be corroborated by evidence in the source code.

Once the hunters return their findings, the validation stage checks the proposed mitigations before assigning each vulnerability an initial risk rating based on source code. The network evidence we collect can raise that rating further when, for example, the affected endpoint carries significant traffic or shows signs of active probing.

The result is a prioritized list of findings, each with a recommended code patch and, when the evidence supports it, a Cloudflare WAF Custom rule that can reduce exposure while the code fix is reviewed. If you have authorized our VDR to defend your zone, we will deploy the rules, scoped conservatively around the method, path, and other request details needed to reach the vulnerable code. If a route pattern contains only variables and wildcards, we do not suggest a rule. We would rather miss a possible connection than claim one the evidence cannot support.

The HTTP method override bypass example above shows how these signals work together. The harness maps the source finding to the production route, uses traffic and security activity to prioritize it, and scopes a proposed WAF rule around the requests that can reach the vulnerable code. That rule can reduce exposure while engineering reviews and ships the code patch.

Where the model runs

When you authorize an investigation, Vulnerability Discovery and Remediation runs the harness on Cloudflare and sends model prompts from Workers through Cloudflare AI Gateway to OpenAI Daybreak models on OpenAI's servers. GPT-5.6 Cyber is used during reconnaissance, hunting, and validation, and its responses return to the harness so the workflow can continue on Cloudflare. No model inference runs at Cloudflare's edge, and the model cannot apply any patch or rule it proposes.

We keep each investigation narrow by limiting it to the source code and evidence the customer authorizes. Before that context reaches the model, Vulnerability Discovery and Remediation removes what the investigation does not need and applies the redaction controls configured for the engagement. The harness treats source code, logs, and request metadata as evidence to inspect, rather than instructions to follow.

Tool access follows the same boundary: each call is logged and checked against the investigation's access policy before it runs, and every patch or rule proposal must pass checks implemented outside the model. If one of those checks fails, the workflow stops before the proposal reaches customer review.

Nothing is presented for review until it has cleared the checks and our team validates the output. For an edge-defense suggestion, that means validating the rule syntax and running it against synthetic fixtures that represent expected requests, rather than against customer traffic. If a check fails or the result remains ambiguous, we hold the output back and route it for diagnosis.

Passing those checks still does not change your environment. After validation by our team, Vulnerability Discovery and Remediation prepares the source code patch and WAF rule.

Join early access

Vulnerability Discovery and Remediation is available to selected customers by invitation during early access through our Managed Defense team. Each engagement starts with one application whose codebase the customer authorizes us to investigate. To connect the findings to production, Vulnerability Discovery and Remediation uses authorized read access to the Web Assets operation inventory, the relevant WAF controls, and Workers Trace Events Logpush where available. The investigation is semi-automated, but you review every result before deciding whether to test or deploy a change.

If you're interested in learning more, talk to your Cloudflare account team.

A few weeks ago, we wrote about Project Glasswing and what we observed when we pointed cyber frontier models at our own code. Since then, we’ve seen that the part of the post that has resonated most deeply is the argument that the architecture around the vulnerability matters more than the speed of the patch.

In the conversations we've had with CISOs and security teams since, the questions have been consistent: what does our architecture actually look like, what should we monitor for, where do we start, and how can Cloudflare help?

Before getting into the details: the architecture below is built almost entirely from Cloudflare's own products, because Cloudflare security is customer zero for the security products we build. The Cloudflare stack already exists in front of our code, employees, and customer-facing applications. If you're a Cloudflare customer, every layer below is available to you today. If you're not, the principles still apply to whatever stack you've built.

What a cyber frontier model actually changes

In the previous post, we showed how a cyber frontier model like Mythos changes the attacker’s timeline. It can find vulnerabilities, reason through exploit chains, and generate working proofs faster than earlier models. While models like Mythos do not change the shape of an intrusion — reconnaissance, initial access, lateral movement, persistence, and exfiltration still have to happen — the difference is in the speed and scale. When pointed at the open web, a model can find and hit low-hanging fruit quickly. Against a hardened target, it still has to probe, and adapt, and it often produces more noise than a careful human operator would.

Discovery, exploit chain construction, and proof-of-concept generation used to be the gating constraints on producing a working attack. A frontier model handles all three in a fraction of the time. Work that used to be slow and methodical is now fast and indiscriminate.

While AI is accelerating how fast developer teams at Cloudflare and many other companies can ship code, the security team’s work has not compressed the same way. An attacker only needs one opening to get in, while security teams need to find and close them all. Writing a fix, regressing it, and shipping it without breaking the code around it has constraints that AI doesn't remove. We learned this the hard way when we let an AI coding assistant write its own patches against our own bugs, as we described at the end of the previous post. Some of those patches fixed the original bug while quietly breaking something else the code depended on.

As these models become more competent and capable, our main focus from a threat standpoint comes down to three things. Each one shapes the architecture we walk through in the rest of this post.

• The first is the speed of discovery. Frontier models make it easier to search large bodies of public code, including the open-source libraries that many companies depend on. That does not mean every bug in a library is exploitable, or that library bugs are where most vulnerabilities live. Exploitability still depends on how the code is used, whether attacker-controlled input can reach the vulnerable path, and the protections that sit around it. But widely used open-source libraries and frameworks give attackers a shared surface to study at scale. When a real, reachable vulnerability exists there, a model can help find it, reason about possible exploit paths, and generate proof-of-concept variants faster than maintainers and defenders can review every downstream use. The gap between when an attacker discovers a vulnerability and when defenders learn it exists is what worries us most. If you are not running these models against your own code, it is safe to assume someone else is.
• The second is exploit volume and adaptation. A model can produce thousands of variations of a single exploit and run reconnaissance at the same scale. All that volume gives an attacker an advantage, but it won’t necessarily get them past signature-based detections. Many of those iterations will have the same underlying signature, so a rule that catches the first one will catch the rest. Adaptation is how they will get past signature-based detections. Ask a model to show you a SQL injection, and it will return a textbook example. Tell it there is a WAF in the way, and it will start probing, learning what gets blocked, and rewriting the payload until it can slip past the rule blocking it.
• The third is the impact when a vulnerability is inevitably exploited. No architecture catches everything. After the vulnerability is exploited, the question we ask ourselves is: where can the attacker get to with one identity, one path, or one credential, before something else stops them? If the answer is "anywhere they want," the vulnerability was never the problem. The architecture around the vulnerability was.

Cloudflare’s superpower: visibility

We see roughly a fifth of the web and that tells us, in real time, which payloads are mutating, which patterns are picking up, and where attacker tooling is moving next. Two teams turn that visibility into defense.

First is Cloudforce One, our threat intelligence, research, and operations team, which sits within the Cloudflare security organization. They turn what we see across the network into insights the rest of the stack can act on: tracked adversaries, emerging campaigns, and indicators of compromise (IOCs). The hard part of this work was never knowing what is malicious — it was the delay in mitigation. Knowledge of a new threat normally has to travel from a threat report, into a feed, and then into a company’s defense before it can be used to block anything. Attackers have learned to move faster than that. Our network closes that gap: Cloudflare customers can now use Cloudforce One threat intelligence directly within the WAF to block high-risk traffic.

Second is the team that owns the WAF engine that does the actual detecting: the managed rulesets that run in front of our own properties and are available to every Cloudflare customer, the machine learning behind WAF Attack Score, and the relationships that sometimes let us ship a rule before a CVE is publicly disclosed. The team is globally distributed and moves fast, releasing rules within hours of a proof-of-concept of an attack becoming known. Once a detection is deployed, it reaches our entire network, along with every Cloudflare customer, in under 30 seconds. React2Shell is a recent example: a managed WAF rule was protecting our own properties, and everyone else's on Cloudflare, hours before the official advisory was published.

The scoring layer, the defenses we put in front of the application, and the containment around the vulnerability all build on what these two teams see. 

Scores over signatures

Signature-based defenses were built for a world where novel exploits were scarce and variations took weeks. Cloudflare's traditional SLA from a fresh proof-of-concept to a live, deployed rule has been 12 hours. With the advent of frontier models, this is not good enough anymore. Detections need to be in place before a CVE is discovered. This is why we layer ML-based detection in front of the traditional signature-based WAF.

The model is trained on a large body of past attack traffic, and it catches new variants of vulnerabilities before they're publicly known. A novel SQL injection or remote code execution chain is almost always a rearrangement of attack shapes the model has seen before, even when the specific exploit is brand new. We run the model on every request and assign a WAF Attack Score between 1 and 99, based on how closely the request resembles those underlying shapes, not against a list of known-bad signatures. The lower the score, the more aggressively we treat the request. That score determines whether we let the request through. We apply a similar scoring methodology to AI prompts with AI Security for Apps: rather than check each prompt against a list of known malicious prompts, we score how closely a prompt resembles an actual attack. 

The architecture around the vulnerability

Those capabilities only matter once they're stacked in front of an application, and the first layer in our defense-in-depth approach is the WAF. Anything that matches a known-bad pattern gets dropped before it reaches the application, which clears the bulk of the obvious traffic and lets the more specialized layers below focus on what's left.

On the API surface, we run a positive security model through API Shield. Instead of trying to anticipate every bad request, we describe what a valid request to each API looks like, either from the API's own definition or learned from our real traffic, and anything that doesn't fit doesn't get through. This neutralizes the advantage of frontier AI models: because we only permit validated traffic, generating thousands of new attack variations fails to bypass the system.

^{Cloudflare’s layered architecture}

Bot Management catches probing traffic on our network before frontier models can build a map. It scores every request on how likely it is to be automated, using the same signals across our whole network: how the client behaves, whether it looks like a real browser, and whether the connection matches a known-bad pattern. An attack only lands if it can find a soft spot. 

Zero Trust Network Access is used for every internal application. The implicit trust of being inside the network is replaced with explicit per-request identity and policy for every employee accessing every tool. The value of this was clear when one of our engineers shipped a misconfigured tool. A flat network would have exposed everything on the same segment, but in our deployment, the exposure stopped at the tool itself. We built Require Access Protection afterwards so newly deployed or misconfigured applications can't be reachable before an access policy is in place.

IdP Federation makes that secure by default posture easier to keep consistent across every Cloudflare account — which becomes even more necessary when more people are shipping internal tools quickly. Instead of asking each team to wire up SSO separately, we configure our identity provider (IdP) once and share it across the organization. New accounts get SSO automatically, recipient-side IdP connections are read-only, and Access policies in each account still evaluate the resulting identity as part of the normal request flow. 

MCP Server Portal gives teams a controlled way to connect AI agents to enterprise systems. Agents access MCP servers that are centrally managed through a single portal, with every action logged. That way when an agent acts on someone's behalf, we know what it did, what it touched, and whether it should have been allowed to. The full picture of how we built it is in our post on enterprise MCP.

AI Gateway runs in front of our internal AI tools the same way AI Security for Apps runs in front of customer-facing AI features, with the same scoring and the same visibility. Inside the company, the visibility piece is more useful than the blocking, because we needed to see what engineers were actually building before we could write meaningful policy on it.

Where your teams can start 

Frontier models can help attackers find vulnerabilities, adapt payloads, and move faster, but they still have to pass through the layered defense you deploy in front of your application. That is where teams should start:

• Put inspection in front of public applications.
• Define what valid API traffic looks like.
• Use bot detection to limit automated probing.
• Require identity and access policy before any internal tool is reachable.

For AI and agentic systems:

• Route model traffic through a gateway.
• Keep agents connected through approved MCP servers.
• Log what they do. 

The goal is to make sure that when one layer misses, the next layer limits what the attacker can see, reach, or change.

That is the point of the architecture around the vulnerability: to limit the scope of an attack. The vulnerability may be what starts the attack, but the architecture determines how far it can go.

How do we know this approach works?

Plenty of security stacks look impenetrable on a whiteboard but fall over in practice. That is why we test ours continuously, both at the perimeter and inside our environment, with our red team involved across both.

At the perimeter, frontier models are one tool we use to test our application security stack as an adaptive attacker. These models sit alongside the rest of our red team and detection workflows including: manual testing, threat intelligence, observed traffic patterns, proof-of-concept analysis, and signals from our own network. Together, those inputs help us decide where to aim testing: newly launched products, recently changed surfaces, and the paths an attacker is most likely to probe first. The most important part is the process that follows. When something gets through, we identify the gap, use the right mix of tools to understand it, write the rule or mitigation, ship the update, and test again to make sure the gap is closed.

Inside the environment, our red team starts from the assumption that the perimeter has already failed. They look at what has changed, where sensitive systems carry risk, and whether one compromised identity, path, or credential can reach farther than it should. When we change the architecture based on what they find, they run the scenario again against the new version to confirm the gap is actually closed.

We confirm that this architecture is working by continuously testing its behavior during failures, rather than relying on the perfection of individual layers.

If your team is working on the same problems and would like to compare notes, reach out to us at security-ai-research@cloudflare.com.

Cloudflare’s Threat Events provides security analysts with a window into the global threat landscape. The platform offers a peek into the immense traffic that Cloudflare processes every day, so you can see in real time which IPs are attacking specific industries or which threat actors are trending globally. However, translating that visibility into active mitigation has often been a manual, reactive process.

Security teams have faced a recurring frustration: knowing that certain IP addresses were associated with specific threat actors (like Tycoon 2FA or RaccoonO365) or had been seen targeting their specific industry in other regions, but they couldn't easily automate the blocking of these high-risk IPs within their own WAF unless they manually configured the rules. 

We are excited to announce a new integration that brings Cloudflare’s vast threat intelligence directly into your WAF engine: you can now write proactive rules using live intelligence data. This means you can add more intelligence context to protect your application against known bad actors — before they even attempt to touch your infrastructure.

By populating specialized fields during the early stages of a request, the WAF can now screen traffic based on:

• Who is attacking by matching specific threat actor names
• Who they are targeting via the industry or country filters to see who the IP has targeted in the past
• What type of attack using enriched threat context, filtering by attack type (DDoS, WAF, cybercrime, etc.) and the timeframe it was last seen

Always-on detection

This new capability is built on the same always-on detection framework we recently introduced for Attack Signature Detection, a system that identifies common attack patterns in real time without requiring pre-configured rules. By separating detection from mitigation, we ensure that threat intelligence is constantly running in the background, enriching your HTTP request analytics with insightful threat metadata before you even decide to take an action.

The primary advantage of an "always-on" model is the elimination of the traditional "log vs. block" trade-off: visibility in log mode, or protection in block mode. That’s because when a rule blocks a request, you lose visibility into how other signatures would have assessed it — insight that could have helped you strengthen your defenses.

If you have a Cloudforce One subscription, these insights appear in your analytics automatically. You can see which threat actors are hitting your site and which industries those IPs usually target, allowing you to verify traffic patterns before "flipping the switch" to block.

These detections execute with negligible latency, ensuring your performance remains lightning-fast while providing the high-confidence data needed to build robust security policies. While this initial release focuses on IP-based matching, we are already looking toward extending these capabilities to JA3 fingerprints and domain-based matching. This will allow you to block malicious traffic even when attackers rotate IPs, by identifying the unique software signatures or malicious destination links they use in their payloads.

New WAF fields

To make this possible, we've exposed the following specific signals directly to the WAF engine:

Example rule expressions

Because a single IP address could be associated with multiple threat actors or targeted industries simultaneously, these fields are represented as arrays. We use the any() function and [*] wildcard to check whether any value within that threat profile matches your criteria:

• Block known DDoS participants targeting your region: any(cf.intel.ip.target_countries[*] == "FR") and any(cf.intel.ip.datasets[*] == "ddos")
• Protect against specific threat actors targeting the Finance sector: any(cf.intel.ip.target_industries[*] == "Banking & Financial Services") and any(cf.intel.ip.attacker_names[*] == "BLACKBASTA")
• Broad protection against specific high-risk origin countries: any(cf.intel.ip.attacker_countries[*] == "IR")

How to use Threat Events data in your workflows

Whether you prefer a UI-driven approach or Infrastructure as Code, these fields are integrated into your existing workflows.

The WAF rule builder (API & Terraform)

For teams that prefer Infrastructure as Code, the new cf.intel fields are fully integrated into the WAF rule builder for WAF custom rules and rate limiting. You can write complex expressions using the same syntax you use today. Because these are standard WAF fields, they are fully supported via the Cloudflare API and Terraform, allowing you to automate threat blocking across your selected domains or even on your whole account.

^{New fields added to the WAF rule builder to allow users to choose the relevant configuration based on the Threat Events indicators.} 

Visibility in Security Analytics

Deployment is only half the battle. All matches triggered by these threat intelligence fields are logged in Security Analytics. You can drill down into your traffic to see exactly which rule was triggered and which specific indicator matched. These enriched logs allow for faster auditing and postmortem analysis when a rule triggers.

^{Threat event matches surface in Security Analytics, with full context and a one-click option to create a custom security rule.}

One-click rule from the Threat Events dashboard

If you are already using the Threat Intelligence Dashboard to investigate trends, you don't have to copy and paste IP lists. You can create Saved Views based on your specific filters, such as "IPs seen attacking the Financial sector in the last seven days." With a single click, you can export these filters directly into a WAF rule.

^{Saved Views now allow users to easily create WAF rules to match the saved view configuration.} 

Global intelligence across our network

Visibility and ease of use are only possible if the underlying engine is fast. How do we handle millions of threat indicators without slowing down your traffic?

These threat intelligence datasets are compressed into a high-performance format and distributed to every single Cloudflare data center globally. When a request hits our network, the Cloudflare WAF performs an O(1) constant-time lookup against these local datasets. This ensures that whether we are checking against ten indicators or ten million, the latency overhead remains effectively zero (measured in microseconds).

Because an IP can be associated with multiple threat vectors, our engine doesn't stop at the first match. It evaluates the set of all signals associated with that IP simultaneously. This ensures that a rule looking for "Attacker = RU" AND "Target Industry = Banking" will trigger correctly by evaluating the intersection of these attributes in a single pass, providing maximum coverage against multi-vector actors without increasing computational complexity.

Ready to get started?

This feature is available today for customers with any active Cloudforce One subscription:

• Cloudforce One Essentials allows customers to access the default datasets in Threat Events, search for indicators, and conduct threat-hunting investigations
• Cloudforce One Advantage allows customers to access our Threat Intelligence Analyst custom insights via requests for information
• Cloudforce One Elite — our most complete package — includes brand protection, a high number of requests for information, and access to all Threat Events datasets

Ready to turn global insights into local defense? Head over to Threat Events or the WAF section of your Cloudflare Dashboard to start building your first Threat Intel rule, or contact your account team to learn more about subscribing to Cloudforce One.

For the last few months, we've been testing a range of security-focused LLMs on our own infrastructure. These LLMs  help identify potential vulnerabilities in our own systems, so we can fix them – and they also show us what attackers are going to be able to do with the latest models.

None of these LLMs has captured more attention than Mythos Preview, from Anthropic. A few weeks ago, we were invited to use Mythos Preview as part of Project Glasswing. We soon pointed it at more than fifty of our own repositories – to see what it would find, and to see how it works.

This post shares what we observed, what the models did well and what they didn't, and how the architecture and process around them needs to change, so they can be used at scale.

What changed with Mythos Preview

Mythos Preview is a real step forward, and it's worth saying that plainly before getting into anything else. We've been running models against our code for a while now, and the jump from what was possible with previous general-purpose frontier models to what Mythos Preview does today is not just a refinement of what came before.

It's a different kind of tool doing a different kind of work, and that makes a clean apples-to-apples comparison to earlier models difficult. So rather than trying to benchmark Mythos Preview against general-purpose frontier models, it's more useful to describe what it can actually do, and two features that stood out across the work we did with Mythos Preview:

• Exploit chain construction - A real attack rarely uses one bug. It chains several small attack primitives together into a working exploit. For instance, it might turn a use-after-free bug into an arbitrary read and write primitive, hijack the control flow, and use return-oriented programming (ROP) chains to take full control over a system. Mythos Preview can take several of these primitives and reason about how to combine them into a working proof. The reasoning it shows along the way looks like the work of a senior researcher rather than the output of an automated scanner.
• Proof generation - Finding a bug and proving it's exploitable are two different things, and Mythos Preview can do both. It writes code that would trigger the suspected bug, compiles that code in a scratch environment, and runs it. If the program does what the model expected, that's the proof. If it doesn't, the model reads the failure, adjusts its hypothesis, and tries again. The loop matters as much as the bugs it finds, because a suspected flaw without a working proof is speculation, and Mythos Preview closes that gap on its own.

Some of what we describe above is not entirely unique to Mythos Preview. When we ran other frontier models through the same harness, they found a fair number of the same underlying bugs, and in some cases they got further than we expected on the reasoning side too. Where they fell short was at the point of stitching the pieces together. A model would identify an interesting bug, write a thoughtful description of why it mattered, and then stop, leaving the actual chain unfinished and the question of exploitability open. What changed with Mythos Preview is that a model can now take those low-severity bugs (which would traditionally sit invisible in a backlog) and chain them into a single, more severe exploit. 

Model refusals in legitimate vulnerability research

The Mythos Preview model provided by Anthropic, as part of Project Glasswing, did not have the additional safeguards that are present in generally available models (like Opus 4.7 or GPT-5.5).

Despite this, the model organically pushes back on certain requests - much like the cyber capabilities that made it useful for vulnerability hunting, the model has its own emergent guardrails that sometimes cause it to push back on legitimate security research requests. But as we found, these organic refusals aren’t consistent - the same task, framed differently or presented in a different context, could produce completely different outcomes as illustrated in the examples below.

^{Example of Mythos Preview pushing back on building a working proof of concept} 

For example, the model initially refused to do vulnerability research on a project, then agreed to perform the same research on the same code after an unrelated change to the project’s environment. Nothing about the code being analyzed had changed.

In another case, the model found and confirmed several serious memory bugs in a codebase, and then refused to write a demonstration exploit. The same request, framed differently, got a different answer, and even the same request can produce different outcomes across runs due to the probabilistic nature of the model. Semantically equivalent tasks can produce opposite outcomes depending on how and when they’re presented to the model.

This matters because while the model’s organic refusals/guardrails are real, they aren’t consistent enough to serve as a complete safety boundary on their own. That’s precisely why any capable cyber frontier model made generally available in the future must include additional safeguards on top of this baseline behavior - making it appropriate for broader use outside of a controlled research context like Project Glasswing.

The signal-to-noise problem

One of the hardest parts of triaging security vulnerabilities is deciding which bugs are real, which are exploitable, and which need fixing now. This was a hard problem even in the pre-AI world. AI vulnerability scanners and AI-generated code have made it worse, and at Cloudflare we've built multiple post-validation stages to deal with it.

Two factors dominate the noise rate:

• Programming language - C and C++ give you direct memory control and, with it, bug classes - buffer overflows, out-of-bounds reads and writes - that memory-safe languages like Rust eliminate at compile time. We saw consistently more false positives from projects written in memory-unsafe languages.
• Model bias - A good human researcher tells you what they found and how confident they are. Models don't. Ask a model to find bugs, and it will find them, whether the code has any or not. Findings come back hedged with "possibly," "potentially," "could in theory," and the hedged findings vastly outnumber the solid ones. That's a reasonable bias for an exploratory tool. It's a ruinous one for a triage queue, where every speculative finding spends human attention and tokens to dismiss, and that cost compounds across thousands of findings.

Mythos Preview represents a clear improvement here, particularly in its ability to chain primitives - combining multiple vulnerabilities into a working proof of concept rather than reporting them in isolation. A finding that arrives with a PoC is a finding you can act on, and it means far less time spent asking "is this even real?"

Our harnesses are deliberately tuned to over-report, so we see more (and miss less), which comes with a lot more noise. But at triage time, Mythos Preview's output has noticeably higher quality: fewer hedged findings, clearer reproduction steps, and less work to reach a fix-or-dismiss decision.

Why pointing a generic coding agent at a repo doesn't work

When we first started AI-assisted vulnerability research last year, our instinct was the obvious one: point a generic coding agent at an arbitrary repository and ask it to discover vulnerabilities. This approach works, in the sense that the model will produce findings, but it doesn't work in producing meaningful coverage of a real codebase and identifying findings of value. There are two main reasons for this:

• Context - Coding agents are tuned for one focused stream of work: building a feature, fixing a bug, writing a refactor. They ingest a lot of source code, hold a single hypothesis at a time, and iterate against it. That's exactly the wrong shape for vulnerability research, which is narrow and parallel by nature. A human researcher picks one specific thing to look at and investigates it thoroughly. That one thing might be a single complex feature, transitions across security boundaries, or a specific vulnerability class like command injections, where attacker input ends up being run as a shell command. Then they do it again, for a different feature, security boundary, or vulnerability class, several thousand times across the codebase. A single agent session (even with subagents) against a hundred-thousand-line repository can cover maybe a tenth of a percent of the surface in a useful way before the model's context window fills up and compaction kicks in - potentially discarding earlier findings that would have mattered.
• Throughput - A single-stream agent does one thing at a time, but real codebases need many hypotheses against many components at once, with the ability to fan out further when something interesting turns up. You can drive a single agent harder, but at some point you stop being limited by the model and start being limited by the shape of the interaction itself. Using the model directly in a coding agent turns out to be fine for manual investigation when a researcher already has a lead and wants a second pair of eyes. However, it's the wrong tool for achieving high coverage. Once we accepted that, we stopped trying to make Mythos Preview do the wrong job and started building the harness around it instead.

What a harness actually fixes

Four lessons came out of running the work at scale, and each one pointed to the need for a harness that manages the overall execution:

• Narrow scope produces better findings - Telling the model "Find vulnerabilities in this repository" makes it wander. Telling it "Look for command injection in this specific function, with this trust boundary above it, here's the architecture document and here's prior coverage of this area" makes it do something much closer to what a researcher would actually do.
• Adversarial review reduces noise - Adding a second agent between the initial finding and the queue - one with a different prompt, a different model, and no ability to generate its own findings - catches a lot of the noise that the first agent would miss if it just checked its own work. It turns out that putting two agents in deliberate disagreement is way more effective than just telling one agent to be careful.
• Splitting the chain across agents produces better reasoning - Asking "Is this code buggy?" and "Can an attacker actually reach this bug from outside the system?" are two different questions, and the model is better at each one when you ask them separately, because each question is narrower than the combined version.
• Parallel narrow tasks beat one exhaustive agent - Coverage improves when many agents work on tightly scoped questions and we deduplicate the results afterward, rather than asking one agent to be exhaustive.

Each of those observations is about model behavior, and put together they describe something that isn't a chat interface anymore. It's a harness that helps you achieve the final outcomes. The first steps to building a harness are simple, as you can ask the model to help, which is what we did. We used Mythos Preview to build on, tailor, and improve our original harnesses to suit its strengths.

An example of what a harness looks like in practice is described below.

Our vulnerability discovery harness

Here's what our vulnerability discovery harness looks like, stage by stage. It was used to scan live code across our runtime, edge data path, protocol stack, control plane, and the open-source projects we depend on.

What this means for security teams

The loudest reaction to Mythos Preview from other security leaders has been about speed - scan faster, patch faster, compress the response cycle. More than one team we have spoken with is now operating under a two-hour SLA from CVE release to patch in production. The instinct is understandable: when the attacker timeline shortens, the defender timeline has to shorten with it. Faster is not going to be enough, and we think a lot of teams are about to spend a lot of time, effort, and money learning that the hard way.

Patching faster does not change the shape of the pipeline that produces the patch. If regression testing takes a day, you cannot get to a two-hour SLA without skipping it, and the bugs you ship when you skip regression testing tend to be worse than the bugs you were trying to patch. We learned a version of this when we tried letting the model write its own patches and watched a few go out that fixed the original bug while quietly breaking something else the code depended on.

The harder question is what the architecture around the vulnerability should look like. The principle is to make exploitation harder for an attacker even when a bug exists, so that the gap between when a vulnerability is disclosed and when it is patched matters less. That means defenses that sit in front of the application and block the bug from being reached. It means designing the application so that a flaw in one part of the code cannot give an attacker access to other parts. It means being able to roll out a fix to every place the code is running at the same moment, rather than waiting on individual teams to deploy it. 

We also recognize this topic cuts both ways. The same capabilities that helped us find bugs in our own code will, in the wrong hands, accelerate the attack side against every application on the Internet. Cloudflare sits in front of millions of those applications, and the architectural principles described above are exactly the ones our products are built to apply on behalf of customers. We will share more on what that means for customers in the weeks ahead.

If your team is doing similar work and would like to compare notes, reach out to us at security-ai-research@cloudflare.com.

Our research with Mythos Preview was conducted in a controlled environment against our own code; every vulnerability surfaced through this work was triaged, validated, and remediated where action was needed under Cloudflare's formal vulnerability management process.

This work was a team effort. Thanks to Albert Pedersen, Craig Strubhart, Dan Jones, Irtefa Fairuz, Martin Schwarzl, and Rohit Chenna Reddy for their contributions to the research, engineering, and analysis behind this blog post.

Security professionals everywhere face a paradox: while more data provides the visibility needed to catch threats, it also makes it harder for humans to process it all and find what's important. When there’s a sudden spike in suspicious traffic, every second counts. But for many security teams — especially lean ones — it’s hard to quickly figure out what’s going on. Finding a root cause means diving into dashboards, filtering logs, and cross-referencing threat feeds. All the data tracking that has happened can be the very thing that slows you down — or worse yet, what buries the threat that you’re looking for. 

Today, we’re excited to announce that we’ve solved that problem. We’ve integrated Cloudy — Cloudflare’s first AI agent — with our security analytics functionality, and we’ve also built a new, conversational interface that Cloudflare users can use to ask questions, refine investigations, and get answers.  With these changes, Cloudy can now help Cloudflare users find the needle in the digital haystack, making security analysis faster and more accessible than ever before.  

Since Cloudy’s launch in March of this year, its adoption has been exciting to watch. Over 54,000 users have tried Cloudy for custom rule creation, and 31% of them have deployed a rule suggested by the agent. For our log explainers in Cloudflare Gateway, Cloudy has been loaded over 30,000  times in just the last month, with 80% of the feedback we received confirming the summaries were insightful. We are excited to empower our users to do even more.

Security analytics dashboards are powerful, but they often require you to know exactly what you're looking for — and the right queries to get there. The new Cloudy chat interface changes this. It is designed for faster root cause analysis (RCA) of traffic anomalies, helping you get from “something’s wrong” to “here’s the fix” in minutes. You can now start with a broad question and narrow it down, just like you would with a human analyst.

For example, you can start an investigation by asking Cloudy to look into a recommendation from Security Analytics.

From there, you can ask follow-up questions to dig deeper:

• "Focus on login endpoints only."

• "What are the top 5 IP addresses involved?"

• "Are any of these IPs known to be malicious?"

This is just the beginning of how Cloudy is transforming security. You can read more about how we’re using Cloudy to bring clarity to another critical security challenge: automating summaries of email detections. This is the same core mission — translating complex security data into clear, actionable insights — but applied to the constant stream of email threats that security teams face every day.

Analyzing your own logs is powerful — but it only shows part of the picture. What if Cloudy could look beyond your own data and into Cloudflare’s global network to identify emerging threats? This is where Cloudforce One's Threat Events platform comes in.

Cloudforce One translates the high-volume attack data observed on the Cloudflare network into real-time, attacker-attributed events relevant to your organization. This platform helps you track adversary activity at scale — including APT infrastructure, cybercrime groups, compromised devices, and volumetric DDoS activity. Threat events provide detailed, context-rich events, including interactive timelines and mappings to attacker TTPs, regions, and targeted verticals. 

We have spent the last few months making Cloudy more powerful by integrating it with the Cloudforce One Threat Events platform.  Cloudy now can offer contextual data about the threats we observe and mitigate across Cloudflare's global network, spanning everything from APT activity and residential proxies to ACH fraud, DDoS attacks, WAF exploits, cybercrime, and compromised devices. This integration empowers our users to quickly understand, prioritize, and act on indicators of compromise (IOCs) based on a vast ocean of real-time threat data. 

Cloudy lets you query this global dataset in a natural language and receive clear, concise answers. For example, imagine asking these questions and getting immediate actionable answers:

• Who is targeting my industry vertical or country?

• What are the most relevant indicators (IPs, JA3/4 hashes, ASNs, domains, URLs, SHA fingerprints) to block right now?

• How has a specific adversary progressed across the cyber kill chain over time?

• What novel new threats are threat actors using that might be used against your network next, and what insights do Cloudflare analysts know about them?

Simply interact with Cloudy in the Cloudflare Dashboard > Security Center > Threat Intelligence, providing your queries in natural language. It can walk you from a single indicator (like an IP address or domain) to the specific threat event Cloudflare observed, and then pivot to other related data — other attacks, related threats, or even other activity from the same actor. 

This cuts through the noise, so you can quickly understand an adversary's actions across the cyber kill chain and MITRE ATT&CK framework, and then block attacks with precise, actionable intelligence. The threat events platform is like an evidence board on the wall that helps you understand threats; Cloudy is like your sidekick that will run down every lead.

Developing this advanced capability for Cloudy was a testament to the agility of Cloudflare's AI ecosystem. We leveraged our Agents SDK running on Workers AI. This allowed for rapid iteration and deployment, ensuring Cloudy could quickly grasp the nuances of threat intelligence and provide highly accurate, contextualized insights. The combination of our massive network telemetry, purpose-built LLM prompts, and the flexibility of Workers AI means Cloudy is not just fast, but also remarkably precise.

And a quick word on what we didn’t do when developing Cloudy: We did not train Cloudy on any Cloudflare customer data. Instead, Cloudy relies on models made publicly available through Workers AI. For more information on Cloudflare’s approach to responsible AI, please see these FAQs.

This is just the next step in Cloudy’s journey. We're working on expanding Cloudy's abilities across the board. This includes intelligent debugging for WAF rules and deeper integrations with Alerts to give you more actionable, contextual notifications. At the same time, we are continuously enriching our threat events datasets and exploring ways for Cloudy to help you visualize complex attacker timelines, campaign overviews, and intricate attack graphs. Our goal remains the same: make Cloudy an indispensable partner in understanding and reacting to the security landscape.

The new chat interface is now available on all plans, and the threat intelligence capabilities are live for Cloudforce One customers. Learn more about Cloudforce One here and reach out for a consultation if you want to go deeper with our experts.

We are observing stealth crawling behavior from Perplexity, an AI-powered answer engine. Although Perplexity initially crawls from their declared user agent, when they are presented with a network block, they appear to obscure their crawling identity in an attempt to circumvent the website’s preferences. We see continued evidence that Perplexity is repeatedly modifying their user agent and changing their source ASNs to hide their crawling activity, as well as ignoring — or sometimes failing to even fetch — robots.txt files.

The Internet as we have known it for the past three decades is rapidly changing, but one thing remains constant: it is built on trust. There are clear preferences that crawlers should be transparent, serve a clear purpose, perform a specific activity, and, most importantly, follow website directives and preferences. Based on Perplexity’s observed behavior, which is incompatible with those preferences, we have de-listed them as a verified bot and added heuristics to our managed rules that block this stealth crawling.

We received complaints from customers who had both disallowed Perplexity crawling activity in their robots.txt files and also created WAF rules to specifically block both of Perplexity’s declared crawlers: PerplexityBot and Perplexity-User. These customers told us that Perplexity was still able to access their content even when they saw its bots successfully blocked. We confirmed that Perplexity’s crawlers were in fact being blocked on the specific pages in question, and then performed several targeted tests to confirm what exact behavior we could observe.

We created multiple brand-new domains, similar to testexample.com and secretexample.com. These domains were newly purchased and had not yet been indexed by any search engine nor made publicly accessible in any discoverable way. We implemented a robots.txt file with directives to stop any respectful bots from accessing any part of a website:  

We conducted an experiment by querying Perplexity AI with questions about these domains, and discovered Perplexity was still providing detailed information regarding the exact content hosted on each of these restricted domains. This response was unexpected, as we had taken all necessary precautions to prevent this data from being retrievable by their crawlers.

Bypassing Robots.txt and undisclosed IPs/User Agents

Our multiple test domains explicitly prohibited all automated access by specifying in robots.txt and had specific WAF rules that blocked crawling from Perplexity’s public crawlers. We observed that Perplexity uses not only their declared user-agent, but also a generic browser intended to impersonate Google Chrome on macOS when their declared crawler was blocked.

Declared	Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; Perplexity-User/1.0; +https://perplexity.ai/perplexity-user)	20-25m daily requests
Stealth	Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36	3-6m daily requests

Both their declared and undeclared crawlers were attempting to access the content for scraping contrary to the web crawling norms as outlined in RFC 9309.

This undeclared crawler utilized multiple IPs not listed in Perplexity’s official IP range, and would rotate through these IPs in response to the restrictive robots.txt policy and block from Cloudflare. In addition to rotating IPs, we observed requests coming from different ASNs in attempts to further evade website blocks. This activity was observed across tens of thousands of domains and millions of requests per day. We were able to fingerprint this crawler using a combination of machine learning and network signals.

An example: 

Of note: when the stealth crawler was successfully blocked, we observed that Perplexity uses other data sources — including other websites — to try to create an answer. However, these answers were less specific and lacked details from the original content, reflecting the fact that the block had been successful. 

In contrast to the behavior described above, the Internet has expressed clear preferences on how good crawlers should behave. All well-intentioned crawlers acting in good faith should:

Be transparent. Identify themselves honestly, using a unique user-agent, a declared list of IP ranges or Web Bot Auth integration, and provide contact information if something goes wrong.

Be well-behaved netizens. Don’t flood sites with excessive traffic, scrape sensitive data, or use stealth tactics to try and dodge detection.

Serve a clear purpose. Whether it’s powering a voice assistant, checking product prices, or making a website more accessible, every bot has a reason to be there. The purpose should be clearly and precisely defined and easy for site owners to look up publicly.

Separate bots for separate activities. Perform each activity from a unique bot. This makes it easy for site owners to decide which activities they want to allow. Don’t force site owners to make an all-or-nothing decision.

Follow the rules. That means checking for and respecting website signals like robots.txt, staying within rate limits, and never bypassing security protections.

More details are outlined in our official Verified Bots Policy Developer Docs.

OpenAI is an example of a leading AI company that follows these best practices. They clearly outline their crawlers and give detailed explanations for each crawler’s purpose. They respect robots.txt and do not try to evade either a robots.txt directive or a network level block. And ChatGPT Agent is signing http requests using the newly proposed open standard Web Bot Auth.

When we ran the same test as outlined above with ChatGPT, we found that ChatGPT-User fetched the robots file and stopped crawling when it was disallowed. We did not observe follow-up crawls from any other user agents or third party bots. When we removed the disallow directive from the robots entry, but presented ChatGPT with a block page, they again stopped crawling, and we saw no additional crawl attempts from other user agents. Both of these demonstrate the appropriate response to website owner preferences.

All the undeclared crawling activity that we observed from Perplexity’s hidden User Agent was scored by our bot management system as a bot and was unable to pass managed challenges. Any bot management customer who has an existing block rule in place is already protected. Customers who don’t want to block traffic can set up rules to challenge requests, giving real humans an opportunity to proceed. Customers with existing challenge rules are already protected. Lastly, we added signature matches for the stealth crawler into our managed rule that blocks AI crawling activity. This rule is available to all customers, including our free customers.  

It's been just over a month since we announced Content Independence Day, giving content creators and publishers more control over how their content is accessed. Today, over two and a half million websites have chosen to completely disallow AI training through our managed robots.txt feature or our managed rule blocking AI Crawlers. Every Cloudflare customer is now able to selectively decide which declared AI crawlers are able to access their content in accordance with their business objectives.

We expected a change in bot and crawler behavior based on these new features, and we expect that the techniques bot operators use to evade detection will continue to evolve. Once this post is live the behavior we saw will almost certainly change, and the methods we use to stop them will keep evolving as well. 

Cloudflare is actively working with technical and policy experts around the world, like the IETF efforts to standardize extensions to robots.txt, to establish clear and measurable principles that well-meaning bot operators should abide by. We think this is an important next step in this quickly evolving space.