Authorization Bypass Through User-Controlled Key vulnerability in ash-project ash allows a user of one tenant to read, overwrite and take over another tenant's records via an upserting create action.
Under attribute multitenancy, Ash.Actions.Create (lib/ash/actions/create/create.ex) and Ash.Actions.Create.Bulk (lib/ash/actions/create/bulk.ex) upserted on the primary key, or on an all_tenants? identity, without scoping the conflicting record to the request's tenant. A caller of a create action with upsert? true that accepts the primary key, supplying a record id from another tenant, updated that record's upsert fields and received all of its attributes in the result. With the default upsert fields the tenant attribute was written too, moving the record into the caller's tenant. Read and update policies do not run for upserts, and tenant-scoped create policies pass because the tenant is the caller's own.
This issue affects ash: from 1.22.0 before 3.37.2.
Incorrect Authorization vulnerability in ash-project ash allows a client to read destination records unrelated to the source record through a through relationship.
When a has_many relationship declared with through: was loaded for a single record with a limit of 1, or an unsorted has_one through relationship was loaded with a limit of 1, lateral_join?/4 in lib/ash/actions/read/relationships.ex chose the path that reads the destination without any filter linking it to the source. The load returned the first destination record by sort, or any other record with an offset. The read still runs the destination's policies but carries the relationship as accessing_from, so a client paging such a field (for example teachers(limit: 1, offset: n) through AshGraphql) can enumerate destination records that policies allow only when reached through that relationship.
This issue affects ash: from 3.24.0 before 3.37.1.
Incorrect Authorization vulnerability in ash-project ash allows an actor to infer data in related records they cannot read via Ash.count/2, Ash.exists/2 and Ash.aggregate/3.
Ash.Actions.Aggregate.run/4 (lib/ash/actions/aggregate.ex) applied only the root resource's read policy before running the aggregate query. The read path also applies each related resource's read policy to filter and sort references that cross a relationship, directly (for example comments.body) or through an aggregate over one, but the aggregate path skipped that step. A caller whose filter or sort reaches these functions, for example through Ash.Query.filter_input/2, an ash_lua script, or an AshAi tool offering count or exists results, can test conditions against related rows hidden from them and recover their existence and attribute values one query at a time. Ash.read/2 and its page counts are not affected.
This issue affects ash: from 2.6.0 before 3.34.6.
CVE-2026-94201 is a high-severity vulnerability in the ash-project's ash library affecting versions from 3.5.1 up to but not including 3.34.3. It arises from unsafe handling of :atom-typed attributes with the unsafe_to_atom? constraint set to true, leading to unbounded atom table growth when attacker-controlled filter values are used. This can exhaust the BEAM atom table and crash the node, causing a denial of service. The issue is fixed by adding a coercion method that avoids interning new atoms during filtering comparisons.
Improperly Controlled Modification of Dynamically-Determined Object Attributes vulnerability in ash-project ash allows a user to set the value of a private action argument on the bulk destroy and bulk update paths.
Action arguments declared with public?: false are meant to be set only by trusted server-side code (for example via Ash.Changeset.set_private_argument/3) and must not be settable from end-user input. CVE-2026-55736 fixed the non-bulk changeset path to strip private arguments from user-supplied parameter maps, but the bulk destroy and bulk update paths were not covered.
Ash.Actions.Destroy.Bulk.base_changeset/5 and Ash.Actions.Update.Bulk.base_changeset/5 match every key in the caller-supplied parameter map against all of the action's arguments with no public? check, then apply the matches to the base changeset. A caller who can submit parameters to a bulk destroy or bulk update action (for example through AshJsonApi, AshGraphql, or a controller that forwards request parameters to Ash.bulk_destroy/4 or Ash.bulk_update/4) can therefore set any private argument of that action, including one referenced by an arg(...) template in the action's changes or validations. Depending on how the application uses the argument (for example an acting_user_id driving authorization or record ownership, or audit metadata), this can lead to an integrity violation or privilege escalation.
The fix requires public? in the argument matching on both bulk paths; private arguments remain settable server-side via the :private_arguments option.
This issue affects ash: from 2.17.15 before 3.33.11.
Ash field_policies are documented to protect against filter-based information disclosure: when a field the actor may not see is referenced in a filter, it is replaced with an expression that evaluates to nil, so a filter cannot be used as a yes/no oracle to read a value the actor cannot see.
This nilling was applied to attributes but not to calculations or aggregates. A user-supplied filter reference to a calculation or aggregate carries an Ash.Query.Calculation / Ash.Query.Aggregate struct, which the authorizer's reference replacement did not match (it only matched the Ash.Resource.* structs), so the filter ran against the real value.
As a result, an actor whose field policies forbid a calculation or aggregate can still filter by it (for example filter(secret_calc == "x") or filter(comment_count == n)) and learn the value from whether rows match — an oracle that recovers field-policy-protected values one probe at a time. Filtering is commonly exposed to lower-privileged actors (for example via AshGraphql or AshJsonApi filter arguments), which is exactly the surface field policies are meant to protect.
The fix routes filter references to calculations and aggregates through the same field-policy nilling as attributes.
This issue affects ash: from 2.11.0-rc.0 before 3.33.4.
CVE-2026-82752 is a medium severity vulnerability in the ash-project ash framework affecting versions from 0.10.0 up to but not including 3.33.0. It involves improper validation of specified quantity in input, where string length constraints are measured by Unicode graphemes rather than codepoints. This discrepancy allows an attacker to store arbitrarily large values in attributes despite small maximum length constraints, potentially leading to unbounded storage growth. The mismatch between validation and storage length counting can also cause data truncation or rejection at the database layer.
Incorrect Authorization vulnerability in ash-project ash returns records that a runtime read policy denies to any actor.
When a resource has an access_type :runtime read policy (a check evaluated per record rather than compiled to a filter), Ash.Policy.Authorizer decides each record in check_result/1 (lib/ash/policy/authorizer/authorizer.ex) by discarding impossible policy scenarios and inspecting what remains. When every scenario for a record was impossible, meaning no policy can authorize it and it must be forbidden, the empty-scenario branch instead kept the record ({[record | data], authorizer, any_forbidden?}) and returned it as authorized. As a result, records the runtime read policy denies are returned to any actor. The fix forbids a record whose scenarios are all impossible.
This issue affects ash: from 3.4.44 before 3.32.2.
Incorrect Authorization vulnerability in ash-project ash widens a relationship's parent(...) scoping filter to match unintended records when the referenced parent field cannot be resolved.
Loading a relationship whose filter references parent(...) resolves that expression against the parent record. resolve_parent_in_filter/3 (lib/ash/actions/read/relationships.ex) resolved an unresolvable parent reference (for example when the referenced field was not selected on the source query) to nil rather than failing. A scoping predicate such as org_id == parent(org_id) then becomes an IS NULL match, and a guard like is_nil(parent(org_id)) or org_id == parent(org_id) activates its unrestricted branch, so the relationship returns records the scope was meant to exclude. The fix fails the read with an error when a parent(...) reference cannot be resolved, instead of defaulting to nil.
This issue affects ash: from 3.13.2 before 3.32.2.
Incorrect Authorization vulnerability in ash-project ash authorizes an aggregate under one read action while computing it under another, so an aggregate can run with policies that do not match the action it was authorized against.
Ash.Actions.Aggregate groups aggregates by their {authorize?, read_action} and authorizes each group under that read action, but when building the data query it selected the action as opts[:action] || read_action || <primary read> (lib/ash/actions/aggregate.ex). When a caller passed an :action option, the aggregate query ran under that action while authorization had been computed for the group's own read_action. If the run action's read policies are more permissive than the authorized one, the aggregate (a count or sum) is computed over records the authorized action's policies would have excluded, disclosing information about data the actor cannot read. The fix runs the aggregate under the same read_action it is authorized against.
This issue affects ash: from 3.5.13 before 3.32.2.
Missing Authorization vulnerability in ash-project ash allows an actor to update records forbidden by resource policies through the atomic path of Ash.update_many/4.
Ash.update_many/4 runs as a single atomic statement (a data-layer update_many, for example a SQL MERGE) whenever an atomic strategy is used and the data layer supports it. Ash.Actions.Update.UpdateMany (lib/ash/actions/update/update_many.ex) took that path even under authorize?: true without applying the resource's policies, so the statement updated every row matched by primary key regardless of the policy filter that authorization would impose. An actor could therefore update records the policies forbid, such as rows belonging to another actor or tenant. The fix restricts the atomic path to data layers supporting changeset filters when authorizing, authorizes each changeset, and merges the resulting policy filter into each changeset so the statement only touches authorized rows.
This issue affects ash: from 3.29.0 before 3.32.2.
Improper Access Control vulnerability in ash-project ash lets a create action overwrite an existing record when the ETS or Mnesia data layer is used, because neither enforced primary-key uniqueness on insert.
Unlike a SQL data layer, whose unique primary-key constraint rejects a duplicate, the ETS and Mnesia data layers implemented create as a keyed insert that replaces any existing entry with the same primary key (lib/ash/data_layer/ets/ets.ex, lib/ash/data_layer/mnesia/mnesia.ex). An actor who can set the primary key on a create (for example a user-supplied string or integer key) can submit a create whose key matches an existing record and silently overwrite it, destroying and replacing another entity's data without going through the update action or its policies. The fix rejects a create whose primary key already exists with an already-taken error, and only allows duplicates for keyless resources.
This issue affects ash: from 0.4.0 before 3.32.2.
Not Failing Securely (Failing Open) vulnerability in ash-project ash skips an Ash.Reactor change when the guard controlling it raises, so a change meant to run does not.
An Ash.Reactor change step can be gated by where validations that decide whether the change runs. Ash.Reactor.ChangeStep (lib/ash/reactor/steps/change_step.ex) evaluated those guards in apply_where_clauses/3, and apply_validation rescued any exception into {:error, error}. The reduce treated that identically to a guard whose condition was simply not met and bypassed the change. So when a guard raises (for example on attacker-influenced input), a change that enforces a security-relevant modification is skipped rather than failing the step. The fix distinguishes a raised exception (now {:raised, error}) and halts the step with an error, failing closed.
This issue affects ash: from 3.0.0-rc.17 before 3.32.2.
Uncontrolled Resource Consumption vulnerability in ash-project ash lets a slow asynchronous read spin a scheduler thread at full CPU while the framework waits for it.
Ash.Actions.Read.AsyncLimiter.await_at_least_one/1 (lib/ash/actions/read/async_limiter.ex) waited for concurrent async read tasks by polling each with Task.yield(task, 0) in a tight loop rather than blocking. While every outstanding task is still running (a slow related-data load or calculation), the loop returns immediately and repeats, busy-spinning and holding a BEAM scheduler at full CPU for the whole duration of the slow read; concurrent slow reads tie up further schedulers. The fix waits with Task.yield_many (a non-blocking sweep followed by a blocking wait with timeout: :infinity), so the process sleeps until a task completes instead of spinning.
This issue affects ash: from 2.19.0 before 3.32.2.
Uncontrolled Resource Consumption vulnerability in ash-project ash lets an attacker exhaust node memory by matching a filter that spans multiple to-many relationships in memory.
Ash.Filter.Runtime matches a filter against an in-memory record by first expanding the record into combinations of its related rows. flatten_relationships/2 (lib/ash/filter/runtime.ex) eagerly built the full Cartesian product across the filter's to-many relationship paths, so a record with K to-many relationships of M rows each materialized on the order of M^K scenarios before any predicate was checked. A filter or dataset that reaches several sizeable to-many relationships therefore allocates memory combinatorially and can exhaust the node. The fix streams the expansion lazily and short-circuits on the first matching scenario, bounding the work.
This issue affects ash: from 1.29.0-rc0 before 3.32.2.
Improper Validation of Specified Type of Input vulnerability in ash-project ash lets an attacker confuse the stored type tag of an Ash.Type.Union value that uses storage: :map_with_tag, bypassing that member's validation and any tag-based authorization.
For a union with storage: :map_with_tag, each member is identified in storage by a configured tag and tag_value. Ash.Type.Union.dump_to_native/2 (lib/ash/type/union.ex) did not force the configured tag when writing the value, so a tag carried in the submitted value was persisted verbatim. An attacker can therefore store a value whose data belongs to one member but whose tag names a different member. On read the value is re-selected by its tag and treated as the incompatible member (a type confusion), bypassing the real member's constraints and any logic or policy that branches on the union tag. The fix drops any incoming tag and forces the configured tag value on dump.
This issue affects ash: from 2.14.18 before 3.32.2.
Improper Input Validation vulnerability in ash-project ash fails to enforce the outer array constraints on a doubly-nested {:array, {:array, type}} attribute, letting invalid input pass validation.
Ash.Type.apply_constraints/3 (lib/ash/type/type.ex) handled the {:array, {:array, type}} case by mapping only the inner {:array, type} constraints over each element, so constraints declared on the outer array (such as min_length, max_length, and nil_items?) were never applied. An attacker could submit an outer list that violates those constraints (too many elements, or nil entries where disallowed) and have it accepted and persisted. The fix enforces the outer array constraints and adds explicit handling for nil and non-list inputs.
This issue affects ash: from 2.16.1 before 3.32.2.
Generation of Error Message Containing Sensitive Information vulnerability in ash-project ash discloses the stored value of a confirmed field to an actor who fails its confirmation check.
Ash.Resource.Validation.Confirm's atomic implementation (atomic/2 in lib/ash/resource/validation/confirm.ex) built the mismatch error with its value set to the field being confirmed. When the actor supplies only the confirmation argument and not the field itself, value resolves through atomic_ref/2 to the field's current stored value, so the mismatch error echoes that stored value back to the actor. Against a confirmation guarding a sensitive attribute, an actor can submit a deliberately wrong confirmation and read the real value from the returned error. The fix reports the actor-supplied confirmation in the error instead of the stored field value.
This issue affects ash: from 2.17.20 before 3.32.2.
Improper Input Validation vulnerability in ash-project ash allows an attacker to persistently deny reads of a record by storing a non-version-7 UUID in an Ash.Type.UUIDv7 attribute.
Ash.Type.UUIDv7.cast_input/2 accepts any well-formed UUID string, including non-version-7 UUIDs, and stores it as a 16-byte binary. On read, cast_stored/2 (lib/ash/type/uuid_v7.ex) routes the stored binary back through cast_input/2, which since an input-validation tightening in v3.6.3 matches only version-7 (and optionally version-4) 16-byte binaries and otherwise expects a 36-character string. A stored non-v7 16-byte binary matches neither clause and returns :error, so every later read of that record fails. An attacker able to set such an attribute poisons the row permanently. The fix decodes any 16-byte stored binary directly in cast_stored/2.
This issue affects ash: from 3.6.3 before 3.32.2.
Integer Overflow or Wraparound vulnerability in ash-project ash lets an attacker corrupt a stored vector and crash later reads of it by submitting a vector with more than 65,535 elements.
Ash.Vector.new/1 (lib/ash/vector.ex) encodes a vector as <<dim::unsigned-16, 0::unsigned-16>> followed by the element floats, packing the element count into a 16-bit field without checking its range. A list of more than 65,535 elements wraps the dimension modulo 65,536, so the encoded header records a dimension that disagrees with the number of stored floats. from_binary/1 later reads binary-size(dim)-unit(32) from the wrapped header, so every read of the corrupted value misparses and raises, denying access to the affected record. The fix rejects any vector whose dimension exceeds 65,535.
This issue affects ash: from 2.14.13 before 3.32.2.