CVE & CISA-KEV Catalog

http.cookies.Morsel.js_output() returns an inline <script> snippet and only escapes " for JavaScript string context. It does not neutralize the HTML parser-sensitive sequence </script> inside the generated script element. Mitigation base64-encodes the cookie value to disallow escaping using cookie value.

Published 2026-06-28

Dragonfly is an in-memory data store built for modern application workloads. Prior to 1.39.9, Dragonfly has a RESP Protocol Injection via Lua redis.error_reply() in EvalSerializer. An authenticated user can inject arbitrary RESP messages into the connection's response stream, potentially causing response desynchronization in connection-pool clients. This vulnerability is fixed in 1.39.9.

Published 2026-06-26

swift-nio-http2's HTTP/2-to-HTTP/1.1 codec did not validate pseudo-header values for control characters before placing them into the translated HTTP/1.1 message. swift-nio-http2 1.44.1 adds validation of all pseudo-header values (:path, :authority, :scheme, :method, and :status) at both the HPACK header validation layer and the HTTP/2-to-HTTP/1.1 translation layer. Requests or responses containing CR, LF, or NUL bytes in any pseudo-header value are now rejected with a connection error. This issue is fixed in swift-nio-http2 1.44.1.

Published 2026-06-25

An attacker sending a large number of crafted DNS queries might be able to trigger a dynamic block being inserted with a value causing invalid output to be produced in the prometheus endpoint. The prometheus endpoint will then be rejected by the scraper until the dynamic block expires.

Published 2026-06-25

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, it does not escape the untrusted fields (name, version, author, description) when they are serialized into the data-obj HTML attribute of each marketplace card. Because the attribute is single-quoted and the value is produced with JSON.stringify() (which does not escape ', <, or >), a package whose name contains a single quote breaks out of the attribute and injects arbitrary HTML. In the desktop client the main BrowserWindow runs with nodeIntegration: true, contextIsolation: false, so the injected markup escalates from DOM XSS to arbitrary OS command execution. This is the same root cause and same impact as the original advisory, reached through a sibling sink the patch did not cover. This vulnerability is fix

Published 2026-06-24

Warp is an agentic development environment. From 0.2024.03.12.08.02.stable_01 until 0.2026.05.06.15.42.stable_01, Warp contains an OS command injection vulnerability in the WSL URL-opening fallback. When Warp is running under WSL and cannot open a URL through wslview, it falls back to a Windows command processor path. A URL controlled through terminal output can reach that fallback when the user opens the link. This vulnerability is fixed in 0.2026.05.06.15.42.stable_01.

Published 2026-06-24

Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, Open WebUI patched SVG XSS in user profile images and webhook profile images but forgot to apply the same fix to model profile images. The ModelMeta class has no validate_profile_image_url field validator, and the model image serving endpoint has no MIME allowlist or nosniff header. Any authenticated user with workspace.models permission (enabled by default) can store a data:image/svg+xml;base64,... payload in a model's profile image and achieve full account takeover of anyone who navigates to the image URL. This vulnerability is fixed in 0.9.6.

Published 2026-06-23

Caddy is an extensible server platform that uses TLS by default. Prior to 2.11.4, Caddy’s stripHTML template function cannot reliably remove all HTML tags from input strings. Certain malformed HTML, such as <<>img src=x onerror=alert()>, can bypass the tag-stripping logic, potentially leaving dangerous content in the output if it is later rendered as HTML. This may allow client-side XSS in cases where untrusted strings are rendered unsafely. This vulnerability is fixed in 2.11.4.

Published 2026-06-23

ImageMagick before 7.1.2-15 and 6.9.13-40 contains a command injection vulnerability in the SVG decoder that allows attackers to inject arbitrary MVG drawing commands. Attackers can craft malicious SVG files with injected Magick Vector Graphics commands that execute during rendering.

Published 2026-06-23

Fabric.js is a Javascript HTML5 canvas library. Prior to 7.4.0, a potential Cross-Site Scripting (XSS) vulnerability exists in Fabric.js due to improper escaping of user-controlled input during SVG serialization via the toSVG() method. Specifically, the color field within the colorStops array of a fabric.Gradient object is not properly escaped when converted into SVG <stop> elements. If an application renders the generated SVG string into the DOM, this may allow an attacker to inject arbitrary HTML/SVG and execute JavaScript in the victim's browser. This vulnerability is fixed in 7.4.0.

Published 2026-06-22

Hono is a Web application framework that provides support for any JavaScript runtime. Prior to 4.12.25, on AWS Lambda, the ALB single-header response and the VPC Lattice v2 response join multiple Set-Cookie headers into one comma-separated value. Because commas also appear inside cookie attributes (for example Expires dates), clients cannot split the value back into individual cookies and silently drop or misparse them. This vulnerability is fixed in 4.12.25.

Published 2026-06-22

Improper escaping of database table names in the CaptureChangeMySQL Processor included with Apache NiFi 1.2.0 through 2.9.0 allows for injecting SQL commands using crafted naming. Manual quoted boundaries added in Apache NiFi 1.8.0 narrowed the scope of potential injection options, but did not cover additional strategies. Apache NiFi installations that do not use the CaptureChangeMySQL Processor are not subject to this vulnerability. Upgrading to Apache NiFi 2.10.0 is the recommended mitigation, which incorporates more robust identifier escaping.

Published 2026-06-22

Stored cross-site scripting in pgAdmin 4's error-rendering and plan-node-rendering paths. Text returned by a PostgreSQL server (ErrorResponse messages, including object names quoted back inside relation-does-not-exist errors and inside EXPLAIN Recheck Cond / Exact Heap Blocks fields) was passed verbatim through html-react-parser at every user-facing sink — the notifier toasts, FormFooterMessage / FormInput help and error areas, FormNote, ModalProvider AlertContent and confirmDelete, ToolErrorView, the Explain visualiser's NodeText panel, the SQL editor confirm dialogs, ConfirmSaveContent, PreferencesHelper modal alerts, and SelectThemes helper text. A PostgreSQL server an attacker controls — or any server returning attacker-influenced text such as a table or column name a low-privilege dat

Published 2026-06-19

HTML injection in pgAdmin 4's cloud deployment module. The verify_credentials, deploy, regions, and update-server endpoints under /rds/, /azure/, /google/, and the top-level /cloud/ blueprint propagated AWS / Azure / Google SDK exception text — and the related file-resolution and database-commit exception text — into the JSON response body (the info and errormsg fields) without HTML-encoding. The Cloud Wizard frontend rendered these strings through html-react-parser, so an attacker-influenced exception message embedded structural HTML directly into the wizard's DOM. The reported entry point is /rds/verify_credentials/. An authenticated pgAdmin user submits a crafted access_key whose value contains an <iframe/src=...> payload; AWS STS rejects the credential with an IncompleteSignature exce

Published 2026-06-19

SQL injection in pgAdmin 4 across every dialog template that renders ``COMMENT ON ... IS '<description>'`` for a user-supplied description field. The Jinja templates for Domains (and their constraints), Foreign Tables, Languages, and Event Triggers, plus the Views OID-lookup query, interpolated the description directly inside a single-quoted SQL literal -- ``'{{ data.description }}'`` -- instead of passing it through the ``qtLiteral`` escape filter. An authenticated pgAdmin user with permission to create or alter the affected object types could submit a description containing an apostrophe, break out of the literal and chain arbitrary SQL. The injected SQL runs under the PostgreSQL role the user is already authenticated as; for a connected role with ``COPY ... TO/FROM PROGRAM`` (typically

Published 2026-06-19

ApostropheCMS is an open-source Node.js content management system. Version 4.29.0 has a stored cross-site scripting vulnerability in the image widget functionality. A user with the Editor role can configure an image widget link to use a javascript: URL payload. Because editors have permission to publish pages, the malicious widget can be published to the live site. When another user, including an administrator or public visitor, clicks the affected image/link, arbitrary JavaScript executes in the victim’s browser. As of time of publication, no known patched versions are available.

Published 2026-06-12

jmespath.php allows users to use JMESPath, software for declaratively specifying how to extract elements from a JSON document, in PHP applications with PHP data structures. Versions prior to 2.9.1 can generate and execute attacker-controlled PHP code when `JmesPath\CompilerRuntime` is used with an attacker-controlled JMESPath expression. The compiler emits parsed JMESPath function names into generated PHP source without sufficient escaping. A crafted expression can cause the generated cache file to contain executable attacker-controlled PHP, which is then loaded by the compiler runtime. The issue is patched in `2.9.1` and later. As a workaround, disable `JP_PHP_COMPILE` and do not use `JmesPath\CompilerRuntime` with attacker-controlled expressions. Use the default `AstRuntime` for untruste

Published 2026-06-12

Quest Bot is an opensource Discord Bot. Prior to version 1.1.6, the latest release suppresses mentions when creating, unbanning, unwarning, kicking, muting, and unmuting, but stored warning reasons are still printed by /warns without mention suppression. A moderator can create a warning with @everyone or @here in the reason, then make the bot later output that reason through /warns, causing a mass ping if the bot has permission. This issue has been patched in version 1.1.6.

Published 2026-06-12

Quest Bot is an opensource modern Discord Bot built for moderation, utilities and support. Prior to version 1.0.5, the latest release suppresses mentions in several moderation commands, but /unban and /unwarn still echo user-controlled reason text in public bot messages without allowedMentions. A moderator can use @everyone or @here in the reason and make the bot send a mass ping. This issue has been patched in version 1.0.5.

Published 2026-06-11

Quest Bot is an opensource modern Discord Bot built for moderation, utilities and support. Prior to version 1.0.4, several moderation commands echo user-controlled reason text in public bot replies without disabling mention parsing. A moderator who does not have permission to mention everyone can still make the bot send @everyone or @here if the bot has that permission. This issue has been patched in version 1.0.4.

Published 2026-06-11

Quest Bot is an opensource modern Discord Bot built for moderation, utilities and support. Prior to version 1.0.3, a normal user can create a ticket with a reason containing @everyone, @here, user mentions, or role mentions. When the ticket is created, the bot posts the attacker-controlled reason into the new ticket channel without suppressing mentions. If the bot has permission to use those mentions, the attacker can make the bot ping staff or everyone with access to the ticket channel. This issue has been patched in version 1.0.3.

Published 2026-06-11

Quest Bot is an opensource modern Discord Bot built for moderation, utilities and support. Prior to version 1.0.3, a normal user can create a reminder whose message contains @everyone or @here. When the reminder triggers, the bot sends the stored message back into the channel without suppressing mass mentions. If the bot has permission to mention everyone, the reminder can ping the entire server or channel later. This issue has been patched in version 1.0.3.

Published 2026-06-11

Xibo is an open source digital signage platform with a web content management system and Windows display player software. Prior to 4.4.2, a vulnerability chain consisting of Stored XSS and Iframe Sandbox escape in the Xibo CMS allows users with DataSet permissions to use the Data Connector functionality to craft messages which escape the sandbox and facilitate XSS. Exploitation of the vulnerability is possible on behalf of an authorized user who has both of the following privileges, which are not granted to non-admins as standard: Include "Add DataSet" button to allow for additional DataSets to be created independently to Layouts Users should upgrade to version 4.4.2 which fixes this issue. Upgrading to a fixed version is necessary to remediate. Users unable to upgrade should revoke such p

Published 2026-06-10

A stored cross-site scripting vulnerability existed in MISP BSimVis tag rendering code. Several client-side rendering paths interpolated tag names, collection names, entity identifiers, cluster names, and tag metadata directly into HTML, HTML attributes, inline JavaScript event handlers, and CSS style values without context-appropriate escaping. The patch adds shared escaping helpers for HTML, attributes, JavaScript strings, and CSS color validation, then applies them across tag badges, tooltips, context menus, cluster cards, autocomplete suggestions, and dynamically inserted tag cards. An attacker able to create or influence stored tag or metadata values could inject a crafted payload that is later rendered in another user’s browser. Successful exploitation could execute arbitrary JavaSc

Published 2026-06-10

FreeSWITCH is a Software Defined Telecom Stack enabling the digital transformation from proprietary telecom switches to a software implementation that runs on any commodity hardware. Prior to version 1.11.0, FreeSWITCH includes a vulnerable function, PREFIX(prologTok)(), in libs/xmlrpc-c/lib/expat/xmltok/xmltok_impl.c, which was cloned from an outdated and vulnerable version in libexpat/libexpat. The function did not receive the corresponding security patch. This issue has been patched in version 1.11.0.

Published 2026-06-09

A YAML injection vulnerability exists in the Windows.Collectors.Remapping artifact of Rapid7 Velociraptor before version 0.76.6. The hostname field in client_info.json inside a collection ZIP is inserted into a YAML template via Go's text/template without escaping. An attacker providing a crafted collection ZIP can leverage literal double quotes and newlines in the hostname to break out of the YAML quoted string and inject a new mount remapping entry. When an analyst applies the generated remapping file with --remap, arbitrary VQL executes on their machine with NullACLManager (all permissions granted, unsandboxed).

Published 2026-06-09

HAX CMS helps manage microsite universe with PHP or NodeJs backends. A stored cross-site scripting (XSS) vulnerability exists in versions prior to 26.0.0 due to improper sanitization of the `<video-player>` component. The component allows `javascript:` URIs in the `source` attribute, which are executed when the page is viewed. This enables attackers to execute arbitrary JavaScript in the context of the victim’s browser and access sensitive data such as JWT tokens and more. Version 26.0.0 fixes the issue.

Published 2026-06-05

A vulnerability in the CLI of Cisco Catalyst SD-WAN Controller, formerly SD-WAN vSmart, Cisco Catalyst SD-WAN Manager, formerly SD-WAN vManage, and Cisco Catalyst SD-WAN Validator, formerly SD-WAN vBond, could allow an authenticated, local attacker to execute arbitrary commands as root by supplying a crafted file to the affected system. This vulnerability is due to insufficient validation of user-supplied input. An attacker could exploit this vulnerability by uploading a crafted file to the affected system. A successful exploit could allow the attacker to perform command injection attacks on an affected system and elevate their privileges as the root user. To exploit this vulnerability, the attacker must have netadmin privileges on the affected system. This would require valid cr

Published 2026-06-04

GLPI is a free asset and IT management software package. Starting in version 10.0.4 and prior to version 10.0.25, a technician can store an XSS payload in the asset locked tab. Upgrade to 10.0.25 or 11.0.7 to receive a patch.

Published 2026-06-03

Improper Encoding or Escaping of Output vulnerability in elixir-tesla tesla allows multipart part header injection via unescaped Content-Disposition parameter values. Tesla.Multipart.part_headers_for_disposition/1 interpolates each disposition parameter as #{k}="#{v}" with no validation of CR (\r), LF (\n), or double-quote characters. The values come verbatim from the caller via Tesla.Multipart.add_field/4 (the name parameter), Tesla.Multipart.add_file/3, and Tesla.Multipart.add_file_content/4 (both the filename parameter and other disposition opts). A " in the value closes the quoted parameter early; a \r\n ends the Content-Disposition header line and starts a new part header (such as a forged Content-Type), or, after a second \r\n, ends the entire part header block and prepends bytes to

Published 2026-06-02

An improper neutralization of user-controllable input in OTRS or ((OTRS)) Community Edition ticket handling allows authenticated attackers to perform reflected cross-site scripting (XSS) attacks via crafted request parameters associated with ticket actions. By injecting malicious JavaScript into manipulated request URLs, attackers can execute arbitrary script code in the context of an authenticated agent session when the crafted link is opened. This issue affects OTRS: * 7.0.x Please note that ((OTRS)) Community Edition 6.x and before are vulnerable. Products based on the ((OTRS)) Community Edition also very likely to be affected

Published 2026-06-01

go-git is an extensible git implementation library written in pure Go. Prior to 5.19.1 and 6.0.0-alpha.4, go-git's SSH transport constructs the remote exec command by wrapping the repository path in single quotes without escaping single quotes embedded inside the path. A repository path containing a single quote can therefore break out of the quoted region in the exec command and be appended as additional shell tokens. This vulnerability is fixed in 5.19.1 and 6.0.0-alpha.4.

Published 2026-05-28

pam_usb provides hardware authentication for Linux using ordinary removable media. Prior to 0.8.7, src/tmux.c reads the user's $TMUX environment variable, splits it on commas, and interpolates the socket-path component directly into a shell command passed to popen(). Because the value is placed inside double-quotes without sanitisation, any value containing " terminates the quoted string and injects arbitrary shell syntax. popen() runs as root inside the PAM stack. This vulnerability is fixed in 0.8.7.

Published 2026-05-27

GuardDog is a CLI tool to identify malicious PyPI packages. From 2.6.0 to 2.9.0, GuardDog includes attacker-controlled filenames, file locations, messages, and code snippets in its default human-readable output without escaping terminal control characters. A malicious package can therefore inject ANSI or OSC escape sequences into analyst terminals or CI logs.

Published 2026-05-27

A vulnerability was detected in NousResearch hermes-agent up to 2026.4.16. The affected element is an unknown function of the component Slack Agent/Mattermost Agent. The manipulation of the argument format_message results in escaping of output. The attack can be executed remotely. The exploit is now public and may be used. The vendor was contacted early about this disclosure but did not respond in any way.

Published 2026-05-24

CryptPad is an end-to-end encrypted collaborative office suite. In versions prior to 2026.2.0, the HTML sanitizer in Diffmarked.js can be bypassed due to incomplete attribute filtering on restricted tags. The sanitizer validates only the src attribute of <iframe>, <video>, and <audio> elements, leaving all other attributes unchecked. As a result, an attacker can inject arbitrary HTML through srcdoc, completely defeating CryptPad's intended bounce sandboxing and enabling link injection or other interactive content within user-controlled documents. The root cause lies in how the sanitizer classifies and enforces tag restrictions: although it defines both forbidden and restricted tag lists, <iframe> is treated as "restricted" rather than "forbidden." Enforcement then inspects only the src att

Published 2026-05-20

CtrlPanel is open-source billing software for hosting providers. Versions 1.1.1 and prior contain a Stored Cross-Site Scripting (XSS) vulnerability exists in the admin role management interface. In app/Http/Controllers/Admin/RoleController.php, the datatable() method interpolates $role->name and $role->color directly into a element's HTML and style attribute without sanitization, and the chained .rawColumns(['actions', 'name']) call instructs DataTables to render the name column as raw HTML, bypassing automatic output escaping. An admin with role creation or edit permissions can inject a payload such as into the name or color fields, which is persisted to the database and executes in the browser of every admin who loads the /admin/roles page. T

Published 2026-05-19

The MCP Registry provides MCP clients with a list of MCP servers, like an app store for MCP servers. Prior to 1.7.7, the public catalogue UI served at GET / (file internal/api/handlers/v0/ui_index.html) is vulnerable to stored cross-site scripting via the server.websiteUrl field of any published server.json. Server-side validation in internal/validators/validators.go (validateWebsiteURL) only checks that the URL parses, is absolute, and uses the https scheme; it does not reject quote characters. Client-side, the value is interpolated into a double-quoted href attribute via innerHTML, using a homegrown escapeHtml helper that performs the standard textContent → innerHTML round-trip. Per the HTML serialisation algorithm, that round-trip encodes only &, <, > and U+00A0 inside text nodes — it d

Published 2026-05-14

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, SiYuan's Bazaar (community marketplace) renders the name and version fields of a package's plugin.json (and the equivalent theme.json / template.json / widget.json / icon.json) into the Settings → Marketplace UI without HTML escaping. The kernel-side helper sanitizePackageDisplayStrings in kernel/bazaar/package.go HTML-escapes only Author, DisplayName, and Description — Name and Version flow through to the renderer raw. The frontend at app/src/config/bazaar.ts substitutes them into HTML template strings via ${item.preferredName} / ${data.name} / v${data.version} and assigns the result to innerHTML. As a consequence, malicious HTML in either field is parsed and executed when a user opens the marketplace tab. This

Published 2026-05-14

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, he tooltip mouseover handler in app/src/block/popover.ts reads aria-label via getAttribute and passes it through decodeURIComponent before assigning to messageElement.innerHTML in app/src/dialog/tooltip.ts:41. The encoder used at the producer side, escapeAriaLabel in app/src/util/escape.ts:19-25, only handles HTML special characters (", ', <, literal <) — it leaves %XX URL-escapes untouched. So a doc title containing %3Cimg src=x onerror=...%3E round-trips through escapeAriaLabel and the HTML attribute layer unmodified. Then decodeURIComponent on the consumer side converts %3C to a literal < character (a real <, NOT a character reference). When that string is assigned to innerHTML, the HTML5 tokenizer enters

Published 2026-05-14

Hono is a Web application framework that provides support for any JavaScript runtime. Prior to 4.12.18, the JSX renderer escapes style attribute object values for HTML but not for CSS. Untrusted input in a style object value or property name can therefore inject additional CSS declarations into the rendered style attribute. The impact is limited to CSS and does not allow JavaScript execution or HTML attribute breakout. This vulnerability is fixed in 4.12.18.

Published 2026-05-13

YetAnotherForum.NET (YAF.NET) is a C# ASP.NET forum. Prior to 4.0.5 and 3.2.12, the thread posting and reply feature accepts user-supplied content via a a post or reply that is stored server-side and later rendered back into the thread page without adequate HTML sanitization or contextual output encoding. This vulnerability is fixed in 4.0.5 and 3.2.12.

Published 2026-05-12

YetAnotherForum.NET (YAF.NET) is a C# ASP.NET forum. Prior to 4.0.5 and 3.2.12, the application's database logger (YAFNET.Core/Logger/DbLogger.cs) captures the incoming request's User-Agent header into a JObject, serializes it with JsonConvert, and stores the result in the EventLog.Description column whenever an event (e.g., an unhandled exception) is logged. The admin event-log page (YetAnotherForum.NET/Pages/Admin/EventLog.cshtml.cs) later deserializes that JSON in FormatStackTrace() and interpolates the UserAgent value directly into an HTML string with no encoding, and the Razor view EventLog.cshtml emits the result through @Html.Raw. This vulnerability is fixed in 4.0.5 and 3.2.12.

Published 2026-05-12

The issue was addressed with improved input validation. This issue is fixed in Safari 26.5, iOS 18.7.9 and iPadOS 18.7.9, iOS 26.5 and iPadOS 26.5, macOS Tahoe 26.5, tvOS 26.5, visionOS 26.5, watchOS 26.5. Processing maliciously crafted web content may prevent Content Security Policy from being enforced.

Published 2026-05-11

If a trusted template author were to write a <script> tag containing an empty 'type' attribute or a 'type' attribute with an ASCII whitespace, the execution of the template would incorrectly escape any data passed into the <script> block.

Published 2026-05-10

Apache Polaris accepts literal `*` characters in namespace and table names. When it later builds temporary S3 access policies for delegated table access, those same characters appear to be reused unescaped in S3 IAM resource patterns and `s3:prefix` conditions. In S3 IAM policy matching, `*` is treated as a wildcard rather than as ordinary text. That means temporary credentials issued for one crafted table can match the storage path of a different table. In private testing against Polaris 1.4.0 using Polaris' AWS S3 temporary- credential path on both MinIO and real AWS S3, credentials returned for crafted tables such as `f*.t1`, `f*.*`, `*.*`, and `foo.*` could reach other tables' S3 locations. The confirmed behavior includes: - reading another table's metadata control file ([Ice

Published 2026-05-04

pretalx is a conference planning tool. Prior to 2026.1.0, an unauthenticated attacker can send arbitrary HTML-rendered emails from a pretalx instance's configured sender address by embedding malformed HTML or markdown link syntax in a user-controlled template placeholder such as the account display name. The most direct vector is the password-reset flow: the attacker registers an account with a malicious name, enters the victim's email address, and triggers a password reset. The resulting email is delivered from the event's legitimate sender address and passes SPF/DKIM/DMARC validation, making it a ready-made phishing vector. This vulnerability is fixed in 2026.1.0.

Published 2026-04-24

Axios is a promise based HTTP client for the browser and Node.js. Prior to 1.15.1 and 0.31.1, the encode() function in lib/helpers/AxiosURLSearchParams.js contains a character mapping (charMap) at line 21 that reverses the safe percent-encoding of null bytes. After encodeURIComponent('\x00') correctly produces the safe sequence %00, the charMap entry '%00': '\x00' converts it back to a raw null byte. Primary impact is limited because the standard axios request flow is not affected. This vulnerability is fixed in 1.15.1 and 0.31.1.

Published 2026-04-24

AnythingLLM is an application that turns pieces of content into context that any LLM can use as references during chatting. Prior to version 1.12.1, AnythingLLM's in-chat markdown renderer has an unsafe custom rule for images that interpolates the markdown image's `alt` text into an HTML `alt="..."` attribute without any HTML encoding. Every call-site in the app wraps `renderMarkdown(...)` with `DOMPurify.sanitize(...)` as defense-in-depth — except the `Chartable` component, which renders chart captions with no sanitization. The chart caption is the natural-language text the LLM emits around a `create-chart` tool call, so any attacker who can influence the LLM's output — most cheaply via indirect prompt injection in a shared workspace document, or directly if they can create a chart record

Published 2026-04-24

PRSD detection denial of service

Published 2026-04-22