axios is vulnerable to read-side prototype-pollution gadgets that can alter request construction when Object.prototype has already been polluted by a separate vulnerability or dependency. In the bodyless method aliases (axios.get(), axios.delete(), axios.head(), axios.options()), inherited data is read via (config || {}).data before config normalization, causing an attacker-controlled body to be sent on requests that did not set one. Additional low-level paths, only reachable when calling exported adapters/helpers (e.g. lib/adapters/http.js, unsafe/helpers/resolveConfig.js) directly with plain configs and no own proxy or paramsSerializer, can inherit polluted proxy values (routing requests through an attacker-controlled proxy) or paramsSerializer values (attacker-controlled URL serialization). These low-level gadgets do not reproduce through normal high-level axios calls on 1.15.2+. The issue is fixed in axios 1.18.0 and 0.33.0.

axios versions >=1.15.2 and <1.18.0 contain prototype-pollution read-side gadgets in Basic auth subfield handling (lib/adapters/http.js and lib/helpers/resolveConfig.js). When an application is already affected by a separate prototype-pollution primitive and makes an axios request with an own auth object that omits the username and/or password properties, axios reads the inherited Object.prototype.username and Object.prototype.password values and uses them to construct an outbound 'Authorization: Basic ...' header. axios itself does not pollute prototypes. The practical impact is outbound request tampering: an attacker who controls the polluted prototype values can inject attacker-chosen Basic auth credentials or replace an existing Authorization header. Credential disclosure is only possible under additional application-specific conditions.

Traefik versions >= v3.7.0 and <= v3.7.7 contain a path traversal vulnerability in the Kubernetes Ingress NGINX provider's RewriteTarget middleware (generated from the nginx.ingress.kubernetes.io/rewrite-target annotation). When an Ingress path uses a regex that captures attacker-controlled text without requiring a path separator (e.g., path /api(.*) with rewrite target /$1), a crafted request such as /api../admin matches the public router, is rewritten to a dot-segment traversal path (/../admin), and is forwarded without post-replacement normalization validation. A backend that normalizes dot segments resolves the path to a protected endpoint (e.g., /admin) reachable only through a separate router secured with BasicAuth, DigestAuth, or ForwardAuth, resulting in route-level authentication bypass. The issue is fixed in v3.7.8.

better-auth versions greater than 1.3.34 and before 1.4.0 contain a vulnerability in the multi-session plugin's /sign-out after-hook, which trusts raw multi-session cookies and forwards extracted values to internalAdapter.deleteSessions without verifying the cookie signature (e.g., via getSignedCookie). An attacker can supply a forged _multi-* cookie to trigger deletion of arbitrary session tokens.

Data::Entropy versions before 0.010 for Perl read remote entropy sources over plain HTTP. The Data::Entropy::RawSource::RandomOrg and Data::Entropy::RawSource::RandomnumbersInfo remote sources are accessed over plain HTTP. The Data::Entropy::RawSource::RandomOrg integrity check trivially matches any non-empty byte string. Any on-path attacker, such as open WiFi, a compromised ISP, captive portal, or a hostile egress proxy substitutes the response and thereby chooses the bytes returned by rand_bits and rand_int for every application that selected one of these sources via with_entropy_source. The _checkbuf method response is equally attacker-controlled, so the retry/sleep behaviour is steerable too.

The Buckaroo Woocommerce Payments Plugin WordPress plugin before 4.9.0 does not perform any capability check or nonce validation on an AJAX action that processes payment capture refunds, allowing any authenticated user, including Subscribers, to trigger refunds against captured orders.

An issue in the parseGoosePayload() function (/goose/goose_receiver.c) of libiec61850 v1.6 allows attackers to bypass authentication via a captured GOOSE frame.

ANDRITZ HIPASE-250 (formerly 250 SCALA) in affected versions stores and transmits user passwords using a reversible format instead of a one-way password hash. This allows an attacker able to read the credential store or capture network traffic to recover all stored passwords.

The Lightbox with PhotoSwipe WordPress plugin before 5.9.0 does not sanitise or escape a link data attribute before rendering it into the image lightbox caption in the browser, allowing users with author-level access and above (who lack the unfiltered_html capability) to store JavaScript that runs when a visitor or administrator opens the lightbox.

ComfyUI is a modular diffusion model GUI, api and backend with a graph/nodes interface. Prior to 0.28.0, get_model_preview in app/model_manager.py joins an unrestricted filename route capture to a selected model directory without a containment check, allowing an unauthenticated remote attacker to use traversal, encoded traversal, absolute paths, or an unbounded path_index to read image-decodable files and enumerate host paths. get_model_preview (app/model_manager.py) built the path with os.path.join(folder, filename) where filename is an unrestricted {filename:.*} route capture. Literal ../, percent-encoded %2e%2e%2f, and absolute paths all escaped the model directory; path_index was also unbounded. The target file is piped through Pillow and re-encoded as WEBP, so disclosure is limited to image-decodable files plus a file-existence/enumeration oracle (and internal-path leakage via path_index errors). This issue is fixed in version 0.28.0.

A flaw was found in yggdrasil-worker-package-manager. A local attacker with existing access to the system could exploit an argument injection vulnerability in the APT backend. This allows specially crafted package names, which begin with a hyphen, to be misinterpreted as command options by apt-get. Successful exploitation could lead to remote code execution (RCE) with root privileges, enabling the attacker to fully compromise the system's integrity, confidentiality, and availability.

LazyOwn RedTeam/APT Framework is an AI-powered C2 and red-team operations framework. Prior to 0.2.154, LazyOwn ships default C2 credentials LazyOwn and LazyOwn in payload.json and core/payload_schema.py and passes them unchanged to lazyc2.py HTTP Basic authentication, allowing any network-reachable attacker who knows the defaults to authenticate to the C2 dashboard with operator-level access. This issue is fixed in 0.2.154.

LazyOwn RedTeam/APT Framework is an AI-powered C2 and red-team operations framework. Prior to 0.2.154, LazyOwn's lazyc2.py registers an unauthenticated Socket.IO input event handler that dispatches data.get('value') to LazyOwnShell.one_cmd, reaching LazyOwnShell.do_cmd and subprocess.call(command, shell=True), allowing unauthenticated remote code execution in the C2 process. This issue is fixed in 0.2.154.

SGLang contains an unauthenticated RCE in /load_lora_adapter_from_tensors via bypass of SafeUnpickler’s incomplete denylist, allowing arbitrary command execution through crafted base64-encoded pickle payloads.

lmsys

Server-Side Request Forgery vulnerability in malach-it Boruta allows an unauthenticated remote attacker to cause the OAuth/OpenID authorization server to issue outbound HTTP requests to attacker-chosen URIs, including internal services and cloud metadata endpoints. Three code paths fetch remote URIs supplied by the requester without sufficient validation of the target. Boruta.Oauth.Request.Base.fetch_unsigned_request/1 in lib/boruta/oauth/request/base.ex dereferences the OAuth request_uri parameter from the authorization request via Finch.build(:get, request_uri) |> Finch.request(OpenIDHttpClient). Boruta.Openid.parse_registration_params/2 in lib/boruta/openid.ex dereferences the jwks_uri supplied in an OpenID Connect dynamic client registration request. Boruta.Ecto.Clients.refresh_jwk_from_jwks_uri/1 in lib/boruta/adapters/ecto/clients.ex later refreshes the stored jwks_uri for an existing client. In all three paths the only validation is that the URI parses with a scheme (and one of the two request_uri clauses does not even restrict the scheme to http or https). The implementations do not require HTTPS, do not enforce a host or IP allowlist, do not reject loopback, private, link-local, or other non-public ranges after DNS resolution, do not cap response size, and do not constrain redirects. An attacker can therefore steer the server's HTTP client at arbitrary network targets reachable from the Boruta host. This issue affects boruta: from 2.3.2 before 2.3.7.

Authentication Bypass by Capture-replay vulnerability in malach-it Boruta allows an attacker who has obtained a previously valid JWT client assertion to authenticate as the issuing OAuth client after the assertion has expired. Boruta accepts JWT-based client authentication (client_secret_jwt and private_key_jwt token endpoint authentication methods) but never enforces that the assertion's exp claim is in the future. The pre-check helper Boruta.Oauth.Request.Base.check_expiration/1 in lib/boruta/oauth/request/base.ex only verifies that an exp claim is present (it pattern-matches on the existence of the key and returns success), and the Joken token configuration used for signature verification, Boruta.Oauth.Authorization.Client.Token.token_config/0 in lib/boruta/oauth/authorization/client.ex, returns an empty map, so Joken's default exp claim validator is not engaged either. Any attacker who obtains a validly-signed client assertion (for example through logs, reverse proxies, browser tooling, or other observability surfaces) can replay it indefinitely to authenticate as the client and obtain access tokens with that client's privileges. This issue affects boruta: from 2.3.0 before 2.3.7.

Date::Manip versions through 6.99 for Perl return corrupted dates via non-ASCII decimal digits that pass the numeric range tests in check. The parse regexes capture year, month and day with the `\d` shorthand, which on a character string matches the whole Unicode decimal digit property `\p{Nd}` and not just `[0-9]`. Date::Manip::Base::check then validates the captured fields with numeric comparisons alone (`$y<1 || $y>9999`, `$m<1 || $m>12`, `$d<1 || $d>$days`), and _parse_check stores the numified fields (`$y+0`). Perl truncates a string at the first character that is not an ASCII digit, so a field whose leading characters are ASCII digits numifies to an in-range prefix and satisfies every test: a year field of three ASCII digits followed by U+0664 ARABIC-INDIC DIGIT FOUR numifies to 202, giving the year 0202, and one non-ASCII digit in the month or day field shifts those fields the same way. The hour, minute and second fields match explicit ASCII character classes (`0?[0-9]`, `[0-5][0-9]`) and do not shift, though a non-ASCII digit in a fractional hour or minute field truncates the fraction. Any caller that passes an untrusted character string to ParseDate() or Date::Manip::Date->parse() can get back a date that differs from the string it parsed, with no parse error. Where the parsed date gates logic such as an expiry check or a retention window, the shift goes unnoticed.

VMware ESX contains an out-of-bounds write vulnerability in the VMXNET3 virtual network adapter. A malicious actor with local administrative privileges on a virtual machine with VMXNET3 virtual network adapter may exploit this issue to execute code on the host. Non VMXNET3 virtual adapters are not affected by this issue.

The Admin and Site Enhancements (ASE) Pro plugin for WordPress is vulnerable to Remote Code Execution in all versions up to, and including, 8.9.0 via the recursive_html function. This is due to the frontend save handler enforces only a publicly emitted nonce with no authentication check, CAPTCHA validation is bypassable by omitting an attacker-supplied key, and repeater row keys from cfgroup[input] are stored verbatim and later spliced into an eval() call in recursive_html without any sanitization or identifier validation. This makes it possible for unauthenticated attackers to execute code on the server. This requires the [post_cf_form] shortcode to be present on at least one publicly accessible page, as the nonce and session ID needed to reach the vulnerable save handler are emitted to unauthenticated visitors by that shortcode.

Inappropriate implementation in SurfaceCapture in Google Chrome prior to 151.0.7922.72 allowed a remote attacker to leak cross-origin data via a crafted HTML page. (Chromium security severity: Low)

google

VaahCMS versions 2.0.0 through 2.3.4 contain a malicious obfuscated JavaScript payload embedded in the Blade template responsible for rendering security OTP emails, allowing remote attackers to execute unauthorized code in any browser that renders the affected email template with JavaScript enabled. The payload establishes a WebSocket connection to a hardcoded command-and-control endpoint, installs a password-field keylogger using MutationObserver to capture dynamically added inputs, scrapes WhatsApp Web DOM content, and accepts remote commands to redirect or overwrite the rendered page.

Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. In versions 1.88.0 up to but not including 1.107.1 and 2.0.0b1 up to but not including 2.5.0, the UI adapters (AG-UI via Agent.to_ag_ui()/AGUIAdapter, and Vercel AI via VercelAIAdapter) use sanitize_messages to strip unresolved ("dangling") client-submitted tool calls from untrusted message history before it reaches the agent, a defense-in-depth default that prevents the agent from executing tool calls the model never emitted. However, the strip anchored to a message index computed before sanitization ran, so when a trailing client message sanitized to empty and was dropped (for example a client system message under the default manage_system_prompt='server'), a preceding assistant response carrying an unresolved tool call became the new tail and was dispatched without inspection. As a result, a remote client could cause a registered, non-approval server tool to run with client-supplied arguments rather than arguments the model produced. The impact is bounded by what the affected tools do and is most significant for applications that gate tool execution in a model-request hook (before_model_request / after_model_request), since a forged call skips the model turn and bypasses that guardrail; approval-gated tools (requires_approval=True) are not auto-executed by this path. This issue has been fixed in versions 1.107.1 and 2.5.0.

pydantic