Search Results (29 CVEs found)

CVE Vendors Products Updated CVSS v3.1
CVE-2026-133429 1 Pydantic 1 Pydantic 2026-11-20 N/A
When validating JSON data into a model allowing extra data, the keys of extra data aren't validated against the key type annotated on __pydantic_extra__, resulting in extra data not matching the declared type. This is the case if you define __pydantic_extra__ as e.g. dict[Annotated[str, Field(pattern=r'^x_')], int]. The validate_json_by_iteration() method (in pydantic-core/src/validators/model_fields.rs), used to validate JSON data into models, only validates the values of extra data, and stores the keys as is. The issue was fixed in commit 4c65dbda9970a488e2ff21c8693fc6f6d70a3358. Impact: validation bypass, as extra data returned by validation doesn't match the declared type. Affected are applications validating untrusted JSON data into models allowing extra data (with extra='allow'), with a key type annotated on __pydantic_extra__ (e.g. dict[Annotated[str, Field(pattern=...)], int]), using Pydantic from 2.13.0 up to (but excluding) 2.14.2. Python mode validation isn't affected.
CVE-2026-133428 1 Pydantic 1 Pydantic 2026-11-20 N/A
Validating an untrusted URL with a long scheme and many hosts into a multi-host URL field (e.g. PostgresDsn) takes quadratic time and memory: an input of about a hundred kilobytes takes around a gigabyte of memory to validate. The parse_multihost_url() function (in pydantic-core/src/validators/url.rs) parses each host separately, prepending the whole URL prefix (including the scheme) to every one of them. As the scheme can be arbitrarily long, and is parsed and stored once per host, both time and memory grow quadratically with the input size. Allowed schemes (e.g. postgres for PostgresDsn) are only checked after parsing, so they don't prevent the issue. The issue was fixed in commit b4028f9d87e34750e53be0ba011d70f3c0fcc94e. Impact: denial of service through validation of untrusted input. Affected are applications validating untrusted strings into multi-host URL fields (PostgresDsn, MongoDsn, NatsDsn, or custom types built on MultiHostUrl).
CVE-2026-133426 1 Pydantic 1 Pydantic 2026-11-20 N/A
Validating an untrusted string in exponent notation into a Fraction field computes the full power of ten without any bound: an input of a few bytes takes seconds to validate. String inputs are passed to the Fraction constructor. For strings in exponent notation (e.g. '1e8000000'), CPython computes the corresponding power of ten as an integer (in Lib/fractions.py). The issue was fixed in commit 69ed168e9ad465eef4c8d81efb152a93b604daf5. For strings in exponent notation, the exponent is now parsed before calling the Fraction constructor, and exponents larger than 4300 (in absolute value) are rejected. Impact: denial of service through validation of untrusted input. Affected are applications validating untrusted strings into Fraction fields (in both JSON and Python mode, including in strict mode), using Pydantic from 2.10.0 up to (but excluding) 2.14.0. From 2.14.0, exponents larger than 4300 (in absolute value) are rejected, matching the maximum number of digits accepted for integers.
CVE-2026-133424 1 Pydantic 1 Pydantic 2026-11-20 5.3 Medium
When extra data is allowed, providing both the alias and the name of an aliased TypedDict or dataclass field makes the unvalidated value (provided under the name) overwrite the validated one (provided under the alias), resulting in validated data not matching the declared type. Extra items are stored in the same dictionary as the fields (the returned dictionary for TypedDict, the instance __dict__ for dataclasses). As only the alias is marked as used, the field name is treated as an extra item, and its unvalidated value overrides the validated one. The TypedDict validator (pydantic-core/src/validators/typed_dict.rs) and the dataclass validator (pydantic-core/src/validators/dataclass.rs) share this logic. The issue was fixed in commit 34491f5328f168f14e3769cd5f0f83c0af0f4778. Impact: validation bypass, as data returned by validation doesn't match the declared type. Affected are applications validating untrusted data into TypedDict or dataclass types with aliased fields and extra data allowed (with extra='allow', or with the extra_items argument for TypedDict), using Pydantic V2 (up to and including 2.14.1). Pydantic models (BaseModel) are not affected, as extra data is stored separately from fields.
CVE-2026-133422 1 Pydantic 1 Pydantic 2026-11-20 N/A
Validating an untrusted list into a field constrained with unique_items=True with Pydantic V1 takes quadratic time: an input of a few hundred kilobytes takes seconds to validate. The ConstrainedList.unique_items_validator validator (in pydantic/types.py) checks each item against all the following ones. For a list with no duplicates (the common, valid case), this results in a quadratic number of comparisons. The issue was fixed in commit 6ec75e57f277a820ec583e601167f25fe8dc7c87. Users relying on the pydantic.v1 namespace in Pydantic V2 should upgrade to 2.14.2 or later. Impact: denial of service through validation of untrusted input. Affected are applications validating untrusted lists into fields constrained with unique_items=True (using conlist() or Field()). The fix only applies to lists of builtin scalar types (int, float, str, bytes, bool and None), for which duplicates can be detected in linear time. Lists of other types (e.g. models or dictionaries) are still checked in quadratic time, as their items aren't guaranteed to be hashable: bound their size using the max_items constraint. Pydantic V2 models are not affected.
CVE-2026-133421 1 Pydantic 1 Pydantic 2026-11-20 N/A
Validating an untrusted ISO 8601 duration string into a timedelta field with Pydantic V1 takes quadratic time: an input of a few dozen kilobytes takes seconds to validate. The iso8601_duration_re regular expression used to parse ISO 8601 durations (in pydantic/datetime_parse.py) contains an unescaped ".", which also matches digits. On a string of digits, the repeated groups overlap and the regex engine backtracks in quadratic time. The issue was fixed in commit 62791e66a9f8f118e465ed760c9e2ede9ae3055e. Users relying on the pydantic.v1 namespace in Pydantic V2 should upgrade to 2.14.2 or later. Impact: denial of service through validation of untrusted input. Affected are applications validating untrusted strings into timedelta fields.
CVE-2026-133505 1 Pydantic 1 Pydantic-ai 2026-11-20 5.9 Medium
When message history is serialized and loaded again, a tool result that contains a JSON object in the serialized shape of one of Pydantic AI's file types is rebuilt as that file. A third party who controls data a tool returns, such as a record that an MCP server or a function tool returns from an external system, can use this to make the application download a URL on its local network, or make the model provider read an s3:// object or uploaded file with the application's credentials. Applications are affected when tool results can contain third-party data, message history containing those results is serialized and loaded again (stored history, history a UI adapter holds on the server, or Temporal, DBOS or Prefect durable execution), and the model is one Pydantic AI downloads files for, one that honors force_download, or one whose provider resolves non-HTTP file URLs or uploaded files. History that a client submits through the Vercel AI or AG-UI adapter in the default mode is not affected. The issue is fixed in pydantic-ai and pydantic-ai-slim 1.107.10 and 2.66.0.
CVE-2026-133161 1 Pydantic 1 Pydantic-ai-harness 2026-11-20 7.5 High
### Summary ui With `strip_ansi=True`, Pydantic AI Harness's `ToolOutputLimits` capability removes ANSI escape sequences from tool output with a pattern that takes time quadratic in the length of a line. A tool result of a few hundred kilobytes, crafted with many unterminated escape sequences, can occupy the event loop for tens of seconds to minutes, delaying every other request and agent run in the process. ### Details This affects applications that set `strip_ansi=True` on `ToolOutputLimits` (named `OverflowingToolOutput` before 0.13.0) and run tools whose output an attacker can influence, such as shell commands, file reads, fetched pages or MCP servers. Stripping runs on the event loop before the capability reduces the output, so its own size limits do not bound it. The assumed attacker can place text in content a tool returns. ### Mitigation Upgrade `pydantic-ai-harness` to 0.70.0 or later. Until then, leave `strip_ansi` at its default, `False`. On Pydantic AI v1, the harness cannot be upgraded past 0.4.0. Upgrade to Pydantic AI v2 and `pydantic-ai-harness` 0.70.0, or leave `strip_ansi` off.
CVE-2026-133276 1 Pydantic 1 Pydantic 2026-11-20 N/A
When extra data is allowed (extra=Extra.allow), providing a __pydantic_initialised__ key when validating a Pydantic V1 dataclass makes it considered as already validated, resulting in field values not being validated. The __init__() method of Pydantic V1 dataclasses stores all the provided keyword arguments in the instance __dict__ before validation, so a __pydantic_initialised__ keyword argument overrides the internal flag used to skip validation of already validated instances. As a result, data returned by validation doesn't match the declared type. Affected are applications validating untrusted data into Pydantic V1 dataclasses allowing extra data. The issue was fixed in Pydantic 1.10.27. Users relying on the pydantic.v1 namespace in Pydantic V2 should upgrade to 2.14.2 or later.
CVE-2026-114386 1 Pydantic 1 Pydantic-ai 2026-10-22 4.8 Medium
Pydantic AI Harness before 0.59.0 fails to enforce UsageLimits.tool_calls_limit on nested tool calls dispatched from run_code in the CodeMode toolset. Users or prompt-injected content can make the sandboxed program invoke tools via run_nested_call() without check_before_tool_call, exceeding operator limits on costly or side-effecting tools.
CVE-2026-107292 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-14 6.4 Medium
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.34.0 until 2.30.0, the Agent.to_web() and clai web development chat server does not validate the Host header, allowing a website visited by a developer to use DNS rebinding to reach a loopback-hosted agent as a same-origin service. The hostile page can read the served UI and submit chat requests that execute agent tools with the local process's privileges and credentials, causing data disclosure or unwanted side effects. Binding to localhost, Origin checks, and CSRF tokens do not prevent the same-origin DNS rebinding path. This issue is fixed in versions 1.107.5 and 2.30.0.
CVE-2026-107287 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-14 6.5 Medium
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.77.0 until 1.107.7 and 2.52.0, the local web_fetch_tool and the WebFetch local fallback can consume excessive CPU and memory during HTML-to-Markdown conversion of attacker-controlled HTML containing deeply nested block elements. Conversion repeatedly reprocesses accumulated text and can greatly expand intermediate output before the returned-content limit is applied, allowing a model-directed fetch to delay other work in the process. Provider-native web fetching is not affected. This issue is fixed in versions 1.107.7 and 2.52.0.
CVE-2026-107286 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-13 7.5 High
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 2.10.0 until 2.53.0, streamed requests made through ConcurrencyLimitedModel or limit_model_concurrency can retain shared concurrency slots because anyio.CapacityLimiter associates an acquired slot with the borrowing task while streaming cleanup can run in a different task. Early stream termination, cancellation, consumer exceptions, or complete stream_text() consumption with debounce_by=0.1 can therefore leave capacity occupied, eventually preventing later requests that share the long-lived limiter from proceeding and causing a denial of service. Agent-level max_concurrency and non-streaming model requests are not affected. This issue is fixed in version 2.53.0.
CVE-2026-107288 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-13 3.7 Low
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.77.0 until 1.107.6 and 2.44.0, the local web_fetch_tool and the WebFetch local fallback compare blocked_domains entries with a URL hostname before both values are normalized to the form used by getaddrinfo. An attacker-influenced model can use an equivalent IDNA spelling, non-ASCII label separator, case variation, or trailing root label that resolves to a blocked host but does not match the configured string, causing the application to fetch that host with its own privileges. allowed_domains fails closed for unmatched spellings, and private-IP and cloud-metadata protections remain effective. This issue is fixed in versions 1.107.6 and 2.44.0.
CVE-2026-107289 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-13 6.8 Medium
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.56.0 until 1.107.6 and 2.44.0, applications that opt attacker-influenced URLs into local network access through FileUrl with force_download='allow-local' or web_fetch_tool with allow_local_urls=True can bypass the cloud-metadata blocklist by appending an IPv6 zone identifier to an IPv6 metadata address. IPv6Address equality and hashing include the zone identifier, so the blocklist comparison fails even though the network stack ignores the zone on a non-link-local destination and reaches the metadata service, potentially exposing cloud IAM credentials. The opt-in settings are disabled by default, and the issue requires an IPv6-enabled environment. This issue is fixed in versions 1.107.6 and 2.44.0.
CVE-2026-107290 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-13 6.5 Medium
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.77.0 until 1.107.6 and 2.44.0, the local web_fetch_tool and the WebFetch local fallback process server-controlled responses with quadratic title extraction, whitespace normalization, and ordered-list numbering. An attacker-controlled page of modest size can therefore block the event loop for an extended period, stalling other agent runs and requests, while unsupported codecs or excessive HTML or JSON nesting can abort an individual run. This issue is fixed in versions 1.107.6 and 2.44.0.
CVE-2026-107291 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-13 4.3 Medium
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 0.3.4 until 1.107.6 and 2.44.0, OpenTelemetry instrumentation configured with InstrumentationSettings(include_content=False) can still export sensitive agent content through exception.message and exception.stacktrace events, error status descriptions, and model_request_parameters containing instructions or the prompted_output_template. The exposed data is available to readers of the configured telemetry backend and can include tool feedback, provider error bodies, runtime instructions, and structured-output templates even though message attributes are redacted. This issue does not grant new access to agent data, and deployments that do not use include_content=False are not affected by the setting bypass. This issue is fixed in versions 1.107.6 and 2.44.0.
CVE-2026-107293 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-13 5.3 Medium
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 0.3.4 until 1.107.4 and 2.27.1, OpenTelemetry instrumentation configured with InstrumentationSettings(include_content=False) can export retry prompts outside tool calls in gen_ai.input.messages and pydantic_ai.all_messages. Agents using NativeOutput, PromptedOutput, or output validators on text output can therefore disclose validation feedback, including invalid model values quoted by that feedback, to readers of the telemetry backend. Tool-call retries and deployments that do not use include_content=False are not affected by this specific path. This issue is fixed in versions 1.107.4 and 2.27.1.
CVE-2026-107295 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-13 7.6 High
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.34.0 until 1.107.4 and 2.28.0, the Agent.to_web() and clai web development chat endpoint has missing request content-type validation. A website visited by a developer can submit a browser-compatible request to a loopback-hosted chat server, causing the served agent to run and execute tools with the privileges and credentials of the local process; client-relayed approval decisions also leave requires_approval=True tools exposed. Binding to localhost does not prevent a browser page from reaching the loopback address. This issue is fixed in versions 1.107.4 and 2.28.0.
CVE-2026-107294 1 Pydantic 2 Pydantic-ai, Pydantic Ai 2026-10-13 6.5 Medium
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.77.0 until 1.107.2 and 2.24.0, web_fetch_tool, the WebFetch local fallback, and remote FileUrl media downloads buffer the complete HTTP response body before enforcing content-size controls. An attacker-influenced URL can stream an arbitrarily large response that exhausts process memory and crashes the worker; affected media types include ImageUrl, DocumentUrl, VideoUrl, and AudioUrl. SSRF protections remain effective, and the impact is limited to availability. This issue is fixed in versions 1.107.2 and 2.24.0.