A reserved start to a cookie's name, __Secure- or __Host-, that supporting browsers enforce. A __Secure- cookie must be set over HTTPS with the Secure attribute; a __Host- cookie must also have Path=/ and no Domain attribute, which pins it to the exact host that set it. A prefix limits how a cookie can be set; it does not encrypt the cookie or decide who may read it once it is sent.
A cookie set with the Partitioned attribute, which a supporting browser stores separately for each top-level site, so an embedded service gets one jar per site that embeds it instead of one shared across all of them. The attribute requires Secure. Partitioning limits cross-site tracking through that cookie; it does not stop an embedded service from recognising a visitor by other means.
A cookie attribute, Strict, Lax or None, that tells the browser whether to send the cookie on requests that start from another site. Browsers treat a cookie with no SameSite as Lax, and SameSite=None is only accepted with Secure. It reduces cross-site request forgery but does not replace a server-side CSRF check, and site here means scheme plus registrable domain, not the exact host.
A domain suffix under which unrelated parties can register their own names, such as com, co.uk or github.io, listed in the Public Suffix List that browsers use to decide where one site ends and another begins. Without it, two strangers' sites on the same hosting domain would look like one site to cookie and same-site rules. The list marks registration boundaries; it says nothing about who owns a name or whether it can be trusted.
The rule that two URLs are same-site only when both the registrable domain and the scheme match, so http://example.com and https://example.com count as different sites for SameSite cookies in browsers that apply it. A site that still moves visitors between HTTP and HTTPS can lose Lax cookies on those steps. It changes which cookies are sent; it does not redirect anything to HTTPS.
A browser check, set with an integrity attribute on a script or stylesheet tag, that refuses the file when its bytes do not match the expected hash. A file loaded from another origin also needs the crossorigin attribute and a CORS response. Covered in CrawlCheck's guide to rolling out a Content Security Policy from report-only to enforced. It protects one fixed version of a file; it cannot be used for a file that is meant to change, and it does not check the scripts that file goes on to load.
The value of an integrity attribute: one or more hash expressions, each an algorithm name (sha256, sha384 or sha512), a hyphen and the base64 digest of the exact file bytes. When several are listed the browser uses the strongest algorithm it supports. It identifies a byte sequence, not a publisher: a hash taken of a tampered file validates that file just as well.
The Content-Security-Policy-Report-Only header: the browser evaluates the policy and reports what it would have blocked, but blocks nothing. It is the safe first stage of a rollout. CrawlCheck reads this header as well as the enforced one when it looks at a site's policy, and its guide covers the move from report-only to enforced. A site with only the report-only header has no Content Security Policy protection yet.
A Content Security Policy source expression such as 'sha256-...' that allows one inline script or style whose exact bytes hash to that value, without allowing every inline block. Changing a single character, whitespace included, breaks the match. Covered in CrawlCheck's guide to rolling out a Content Security Policy from report-only to enforced. It allows that one block; every other inline script still needs its own hash or a nonce.
A script-src keyword that tells the browser to trust any script loaded by a script it already trusts through a nonce or hash, and to ignore host allowlists and 'self' in that directive. It is what makes nonce-based policies workable with tag managers and other loaders. Covered in CrawlCheck's guide to rolling out a Content Security Policy from report-only to enforced. Trust passes to whatever a trusted script loads, so a compromised trusted script can still bring in more.
A Content Security Policy directive that limits the addresses forms on a page may submit to. It does not fall back to default-src, so a policy without it leaves form targets unrestricted. Covered in CrawlCheck's guide to rolling out a Content Security Policy from report-only to enforced. It governs form submissions only; it does not stop a script sending the same data with fetch, which connect-src controls.
A Content Security Policy directive that makes the browser fetch a page's http:// sub-resources over https:// instead. It clears mixed-content warnings on older pages without editing every link. Covered in CrawlCheck's guide to rolling out a Content Security Policy from report-only to enforced. It only works where the HTTPS version of each file exists, and it does not redirect the page itself or replace HSTS.
A Content Security Policy directive that limits what a page's base element may set as the base for its relative links and script paths. It does not fall back to default-src, so it has to be set on its own, usually to 'self' or 'none'. Covered in CrawlCheck's guide to rolling out a Content Security Policy from report-only to enforced. It protects how relative URLs resolve; it does not restrict where scripts may load from.
A Content Security Policy directive for plugin content loaded by object and embed elements, usually set to 'none' now that browsers no longer run plugins such as Flash. It falls back to default-src when absent. Covered in CrawlCheck's guide to rolling out a Content Security Policy from report-only to enforced. It covers those two elements only; frames are governed by frame-src.
When the Referer header carries a page's full address, path and query string included, to another site the page links to or loads from, exposing tokens, email addresses or search terms in the URL. Browsers now default to sending only the origin cross-site, but a page or link can loosen that; the Referrer-Policy header sets the rule. It controls the header only; it does not hide URL data that a third-party script reads from the page directly.
A DNS CNAME record that still points a subdomain at an outside service, such as hosting, storage or a CDN, after the resource there was deleted. Whoever can claim that name on the service next controls what the subdomain serves. A CNAME pointing elsewhere is normal; it is dangling only when the target no longer belongs to you, which the DNS record alone cannot show.
Taking control of a site's subdomain by claiming the external resource that a dangling DNS record points at, then serving content, reading cookies scoped to the parent domain, or passing as the real brand in phishing. The fix is removing the record in the same change that removes the service. It needs both the dangling record and an unclaimed target; a subdomain served from your own servers cannot be taken this way.
The SPF rule (RFC 7208) that checking a record may trigger at most 10 mechanisms that need DNS lookups, such as include, a, mx, exists and redirect. Past that the result is a permanent error, which receivers treat as failing SPF, and each include of a mail provider can add lookups of its own. Staying under the limit makes SPF checkable; it does not prove the listed senders are the right ones.
A connection made with an older, weaker TLS version or cipher than both sides support, by attack or by misconfiguration. TLS 1.3 includes signals that let a client detect a forced downgrade, and TLS 1.0 and 1.1 are deprecated (RFC 8996). Refusing old versions protects the connection; it says nothing about the certificate or about the site behind it.
Public append-only logs in which certificate authorities record the TLS certificates they issue; major browsers require proof of logging before they trust a public certificate. Searching the logs for a domain shows every certificate issued for it, including ones its owner did not request. The logs show that a certificate exists; they do not revoke it or prove who asked for it.
An HTTP response header that tells a browser whether the page may be shown inside a frame on another site: DENY forbids it everywhere, SAMEORIGIN only on the same origin. CrawlCheck sends DENY on its own sensitive pages, and its broken-embed check reads the header on every frame source a scanned homepage loads. It only matters to browsers that honour it, and a Content-Security-Policy frame-ancestors list overrides it.
The Content-Security-Policy directive that lists which origins may show a page inside a frame, replacing the older X-Frame-Options header. CrawlCheck uses frame-ancestors 'none' on its own sensitive pages and, when checking a site's embeds, treats a list that leaves the page's origin out as a frame the visitor will not see. It controls framing only; it says nothing about whether the framed content is correct or current.
An HTTP response header that sets how much of a page's address a browser passes on as the Referer when a visitor follows a link or loads a resource from it. CrawlCheck sends strict-origin-when-cross-origin on its pages and no-referrer on its private account pages and on redirects that carry an order result in the address. It limits what the browser sends; it does not hide the visit from the site being linked to.
An HTTP response header that switches browser features, such as the camera, microphone, location or payment, on or off for a page and anything it frames. CrawlCheck sends one on its own pages that turns off the features it never uses. It restricts what scripts on the page may ask the browser for; it does not stop a script from running or from sending data it already has.
An extension to TLS that encrypts the first message of a secure connection, including the name of the site being visited, so the network path cannot read it. CrawlCheck notes whether a site's DNS HTTPS record publishes ECH keys. It hides the site name from on-path observers; it does not hide the server's IP address, and it does nothing for the page's content once it arrives.
A domain built to be mistaken for a well-known one: a letter swapped for a look-alike from another script, a one-letter typo, or the brand plus a word like login or support. A match is only a lookalike until other evidence corroborates it, such as a recent registration or a homepage copying the brand. Brand owners often register lookalikes defensively, and those redirect to the real site.
A character that looks identical or nearly identical to another, usually from a different script, such as the Cyrillic а and the Latin a. Homoglyphs let a domain read as a famous name while being a different address. Browsers often show such names in their xn-- punycode form; agents reading raw text do not get that warning.
A control the verified owner of a domain uses during a compromise to tell every agent that connecting to or transacting with the domain is blocked until the owner clears it. It is published as a signed state that reaches agents within a minute and voids receipts issued before it. It is a statement by the owner; it does not repair the compromise.
A part of a site whose change could redirect money, logins or data: the payment processors on the checkout, where the checkout actually goes, crypto addresses printed on the page, the OAuth server, mail servers and MCP endpoints. A change in one of them is held for the owner to confirm before agents transact. A change is not proof of an attack; most are legitimate, which is why the owner confirms.
Malicious instructions embedded in content an agent retrieves or opens rather than typed by the user. Treating remote content as data, not authority, is the core boundary.
User-supplied text attempting to override application or system instructions. Refusal behavior helps but is not a complete authorization control.
A prompt or interaction pattern designed to bypass model restrictions. Success against one version may fail against another, so jailbreak rates require dated and versioned tests.
Unauthorized transfer of secrets or sensitive information from a system. A model can facilitate it through tools, outputs, encoded text, or attacker-controlled destinations.
Exposure of private, confidential, regulated, or security-relevant data in model output or logs. Correctly answering a prompt can still be a disclosure failure.
Accidental exposure of credentials, keys, tokens, or internal values. Redaction at display time does not remove secrets already sent to a provider or log.
Passing model-generated content into interpreters, browsers, databases, or shells without appropriate validation. The output is untrusted even when it came from a trusted model endpoint.
Granting an agent more tools, permissions, autonomy, or action scope than needed. Accuracy improvements do not compensate for avoidable blast radius.
Giving each component only the permissions required for its current task. It limits damage but requires continuous maintenance as tools and workflows change.
The point where data or control moves between components with different assurance levels. Every crossing needs explicit validation rather than inherited trust.
Checking supplied values against allowed types, formats, ranges, and business rules. Schema validation is one layer and cannot detect every harmful valid value.
Checking generated or tool-returned data before display, storage, or execution. It should be tailored to the downstream sink, not treated as generic sanitization.
An explicit set of permitted values, operations, hosts, or paths. It reduces ambiguity but becomes dangerous when broad wildcards silently expand it.
A set of known prohibited patterns or targets. It blocks recognized cases and predictably misses novel encodings, variants, and attack paths.
Server-side request forgery, where attacker-controlled input makes a server request unintended internal or external resources. URL parsing, redirects, DNS changes, and metadata endpoints all matter.
Remote code execution, where an attacker causes code to run on another system. An agent with a shell tool can create equivalent impact without exploiting a traditional software bug.
The resources and actions a credential authorizes. A short-lived credential can still be overpowered if its scope is broad.
A named permission requested and granted within an OAuth authorization flow. The label’s actual force depends on the resource server enforcing it correctly.
A shared secret used to identify or authorize API access. It usually identifies an application rather than a person and should not be embedded in public client code.
A credential accepted from whoever possesses it. Transport security, storage, expiration, and audience restrictions are therefore critical.
Mutual TLS, where client and server authenticate with certificates during connection establishment. It verifies certificate possession, not whether an application request is safe.
Rules controlling how much traffic a principal may send during a period. It protects capacity but can penalize shared addresses and bursty legitimate work.
A record of security-relevant identities, actions, targets, outcomes, and times. Logs support investigation only when complete, protected, retained, and attributable.
Personally identifiable information that can identify or reasonably link to a person. Definitions vary by jurisdiction and context, so one static field list is insufficient.
Collecting, sending, and retaining only data necessary for a stated purpose. It reduces exposure but requires knowing which later uses are genuinely necessary.
Restricting data use to specified, legitimate purposes. Broad consent or vague product improvement language weakens the practical boundary.
The time data remains stored before deletion or anonymization. A stated period is not evidence that every cache, backup, and derived record follows it.
The geographic location where data is stored or processed. Residency does not by itself determine jurisdiction, access, ownership, or privacy compliance.
Controls preventing one customer’s data or actions from crossing into another customer’s environment. Shared models, logs, caches, and vector indexes are all possible boundaries.
A structured account of assets, actors, entry points, trust boundaries, and plausible attacks. It guides controls but becomes stale as architecture and capabilities change.
A DNS record telling receiving mail servers what to do with a message that claims your domain and fails authentication: none, quarantine or reject. It evaluates SPF and DKIM results for alignment with the visible From domain; it authenticates nothing itself, and p=none is monitoring, not protection.
The requirement that the domain which passed SPF or DKIM be the same organisational domain as the visible From address. A sender can pass SPF or DKIM on its own domain and still fail DMARC because nothing aligned, which is how a marketing platform or form tool that signs with its own domain lands in spam under quarantine.
Sender Policy Framework, a DNS record listing the servers allowed to send mail for a domain, checked against the envelope sender's address. It breaks on any forwarding hop, because the forwarder is not in the list, which is why SPF alone cannot make a quarantine policy safe.
DomainKeys Identified Mail, a signature over a message's headers and body made with a private key whose public half is published in DNS under a selector. It survives forwarding because it is computed over the content, and its d= domain is what DMARC aligns against; a domain with mail and no DKIM has one point of failure.
A policy, published at a well-known URL and announced in DNS, telling sending servers that mail to this domain must use TLS to the named mail exchangers. It closes the downgrade attack that plain SMTP allows; testing mode reports, enforce mode refuses, and TLS-RPT is where the reports go.
A DNS record naming an address that receives daily reports of TLS failures when other servers deliver mail to your domain. It is how an MTA-STS policy is monitored before it is enforced; the address has to exist, or the reports bounce into nothing.
A DNS record naming which certificate authorities may issue certificates for a domain. An authority that is not listed must refuse, which limits the damage of a compromised validation path; it does not affect certificates already issued and does nothing if no record exists.
Signatures over DNS records, chained from the root through the registry to the zone, so that a resolver can detect a forged answer. It protects the lookup, not the site: a signed zone can still point at a compromised server. It is switched on at the DNS host and completed at the registrar, and a broken chain makes the domain unreachable.
A permanent delivery failure: the mailbox does not exist. Sending systems respond by suppressing the address so they stop hammering it, which means every later message is accepted by the sending API and dropped, with no error, until the suppression is cleared. It is the failure that looks like success from the sending side.