OpenAI Admin Key Detector (#4689)

* restrict openai detector regex to skip admin key

* add openai admin key detector

* fix: add missing hyphen in openai detector regex

* incorporated feedback
added tests to check exclusivity
tightened regexes

* remove redundant comment

* uniformity in openai regex

Author: amanfcp

pkg/detectors/openai/openai.go

@@ -28,7 +28,10 @@ var (
 	defaultClient = common.SaneHttpClient()
 
 	// The magic string T3BlbkFJ is the base64-encoded string: OpenAI
-	keyPat = regexp.MustCompile(`\b(sk-[[:alnum:]_-]+T3BlbkFJ[[:alnum:]_-]+)\b`)
+	// Matches: legacy keys (sk-{alnum}T3BlbkFJ...), project keys (sk-proj-...),
+	//          service account keys (sk-svcacct-... or sk-service-...)
+	// Does NOT match: admin keys (sk-admin-...)
+	keyPat = regexp.MustCompile(`\b(sk-(?:(?:proj|svcacct|service)-[A-Za-z0-9_-]+|[a-zA-Z0-9]+)T3BlbkFJ[A-Za-z0-9_-]+)\b`)
 )
 
 // Keywords are used for efficiently pre-filtering chunks.

pkg/detectors/openai/openai_integration_test.go

@@ -17,7 +17,7 @@ import (
 )
 
 func TestOpenAI_FromChunk(t *testing.T) {
-	ctx, cancel := context.WithTimeout(context.Background(), time.Second*5)
+	ctx, cancel := context.WithTimeout(context.Background(), time.Second*10)
 	defer cancel()
 	testSecrets, err := common.GetSecret(ctx, "trufflehog-testing", "detectors4")
 	if err != nil {

pkg/detectors/openai/openai_test.go

@@ -8,6 +8,20 @@ import (
 	"testing"
 )
 
+func TestOpenAI_DoesNotMatchAdminKeys(t *testing.T) {
+	d := Scanner{}
+	adminKey := `OPENAI_ADMIN_KEY = "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"`
+
+	results, err := d.FromData(context.Background(), false, []byte(adminKey))
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+
+	if len(results) != 0 {
+		t.Errorf("openai detector should not match admin keys, but got %d results", len(results))
+	}
+}
+
 func TestOpenAI_Pattern(t *testing.T) {
 	d := Scanner{}
 	ahoCorasickCore := ahocorasick.NewAhoCorasickCore([]detectors.Detector{d})

pkg/detectors/openaiadmin/openaiadmin.go

@@ -0,0 +1,111 @@
+package openaiadmin
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"net/http"
+
+	regexp "github.com/wasilibs/go-re2"
+
+	"github.com/trufflesecurity/trufflehog/v3/pkg/common"
+	"github.com/trufflesecurity/trufflehog/v3/pkg/detectors"
+	"github.com/trufflesecurity/trufflehog/v3/pkg/pb/detectorspb"
+)
+
+type Scanner struct {
+	client *http.Client
+}
+
+// Ensure the Scanner satisfies the interface at compile time.
+var _ detectors.Detector = (*Scanner)(nil)
+
+var (
+	defaultClient = common.SaneHttpClient()
+
+	// Admin keys follow the format: sk-admin-{58 chars}T3BlbkFJ{58 chars}
+	// Total length: 133 chars (9 char prefix + 124 chars for key)
+	// where T3BlbkFJ is the base64-encoded string: OpenAI
+	keyPat = regexp.MustCompile(`\b(sk-admin-[A-Za-z0-9_-]{58}T3BlbkFJ[A-Za-z0-9_-]{58})\b`)
+)
+
+// Keywords are used for efficiently pre-filtering chunks.
+func (s Scanner) Keywords() []string {
+	return []string{"sk-admin-"}
+}
+
+// FromData will find and optionally verify Openaiadmin secrets in a given set of bytes.
+func (s Scanner) FromData(ctx context.Context, verify bool, data []byte) (results []detectors.Result, err error) {
+	dataStr := string(data)
+
+	uniqueMatches := make(map[string]struct{})
+	for _, match := range keyPat.FindAllStringSubmatch(dataStr, -1) {
+		uniqueMatches[match[1]] = struct{}{}
+	}
+
+	for token := range uniqueMatches {
+		s1 := detectors.Result{
+			DetectorType: detectorspb.DetectorType_OpenAIAdmin,
+			Redacted:     token[:11] + "..." + token[len(token)-4:],
+			Raw:          []byte(token),
+		}
+
+		if verify {
+			client := s.client
+			if client == nil {
+				client = defaultClient
+			}
+
+			isVerified, verificationErr := verifyMatch(ctx, client, token)
+			s1.Verified = isVerified
+			s1.SetVerificationError(verificationErr, token)
+			s1.AnalysisInfo = map[string]string{
+				"key": token,
+			}
+		}
+
+		results = append(results, s1)
+	}
+
+	return
+}
+
+func verifyMatch(ctx context.Context, client *http.Client, token string) (bool, error) {
+	// Use the Admin API Keys list endpoint to verify the admin key
+	// https://platform.openai.com/docs/api-reference/admin-api-keys/list
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, "https://api.openai.com/v1/organization/admin_api_keys", http.NoBody)
+	if err != nil {
+		return false, err
+	}
+
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", token))
+
+	res, err := client.Do(req)
+	if err != nil {
+		return false, err
+	}
+	defer func() {
+		_, _ = io.Copy(io.Discard, res.Body)
+		_ = res.Body.Close()
+	}()
+
+	switch res.StatusCode {
+	case http.StatusOK:
+		return true, nil
+	case http.StatusUnauthorized:
+		// Invalid admin key - determinate failure
+		return false, nil
+	default:
+		// Unexpected response - indeterminate failure
+		return false, fmt.Errorf("unexpected HTTP response status %d", res.StatusCode)
+	}
+}
+
+func (s Scanner) Type() detectorspb.DetectorType {
+	return detectorspb.DetectorType_OpenAIAdmin
+}
+
+func (s Scanner) Description() string {
+	return "OpenAI Admin API keys provide administrative access to OpenAI organization resources. These keys can be used to manage API keys, audit logs, and other organization-level settings."
+}

pkg/detectors/openaiadmin/openaiadmin_integration_test.go

@@ -0,0 +1,162 @@
+//go:build detectors
+// +build detectors
+
+package openaiadmin
+
+import (
+	"context"
+	"fmt"
+	"testing"
+	"time"
+
+	"github.com/google/go-cmp/cmp"
+	"github.com/google/go-cmp/cmp/cmpopts"
+
+	"github.com/trufflesecurity/trufflehog/v3/pkg/common"
+	"github.com/trufflesecurity/trufflehog/v3/pkg/detectors"
+	"github.com/trufflesecurity/trufflehog/v3/pkg/pb/detectorspb"
+)
+
+func TestOpenAIAdmin_FromChunk(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), time.Second*10)
+	defer cancel()
+	testSecrets, err := common.GetSecret(ctx, "trufflehog-testing", "detectors6")
+	if err != nil {
+		t.Fatalf("could not get test secrets from GCP: %s", err)
+	}
+	secret := testSecrets.MustGetField("OPENAI_ADMIN")
+	inactiveSecret := testSecrets.MustGetField("OPENAI_ADMIN_INACTIVE")
+
+	type args struct {
+		ctx    context.Context
+		data   []byte
+		verify bool
+	}
+	tests := []struct {
+		name                string
+		s                   Scanner
+		args                args
+		want                []detectors.Result
+		wantErr             bool
+		wantVerificationErr bool
+	}{
+		{
+			name: "found, verified",
+			s:    Scanner{},
+			args: args{
+				ctx:    context.Background(),
+				data:   []byte(fmt.Sprintf("You can find an OpenAI admin secret %s within", secret)),
+				verify: true,
+			},
+			want: []detectors.Result{
+				{
+					DetectorType: detectorspb.DetectorType_OpenAIAdmin,
+					Verified:     true,
+				},
+			},
+			wantErr:             false,
+			wantVerificationErr: false,
+		},
+		{
+			name: "found, unverified",
+			s:    Scanner{},
+			args: args{
+				ctx:    context.Background(),
+				data:   []byte(fmt.Sprintf("You can find an OpenAI admin secret %s within but not valid", inactiveSecret)), // the secret would satisfy the regex but not pass validation
+				verify: true,
+			},
+			want: []detectors.Result{
+				{
+					DetectorType: detectorspb.DetectorType_OpenAIAdmin,
+					Verified:     false,
+				},
+			},
+			wantErr:             false,
+			wantVerificationErr: false,
+		},
+		{
+			name: "not found",
+			s:    Scanner{},
+			args: args{
+				ctx:    context.Background(),
+				data:   []byte("You cannot find the secret within"),
+				verify: true,
+			},
+			want:                nil,
+			wantErr:             false,
+			wantVerificationErr: false,
+		},
+		{
+			name: "found, would be verified if not for timeout",
+			s:    Scanner{client: common.SaneHttpClientTimeOut(1 * time.Microsecond)},
+			args: args{
+				ctx:    context.Background(),
+				data:   []byte(fmt.Sprintf("You can find an OpenAI admin secret %s within", secret)),
+				verify: true,
+			},
+			want: []detectors.Result{
+				{
+					DetectorType: detectorspb.DetectorType_OpenAIAdmin,
+					Verified:     false,
+				},
+			},
+			wantErr:             false,
+			wantVerificationErr: true,
+		},
+		{
+			name: "found, verified but unexpected api surface",
+			s:    Scanner{client: common.ConstantResponseHttpClient(404, "")},
+			args: args{
+				ctx:    context.Background(),
+				data:   []byte(fmt.Sprintf("You can find an OpenAI admin secret %s within", secret)),
+				verify: true,
+			},
+			want: []detectors.Result{
+				{
+					DetectorType: detectorspb.DetectorType_OpenAIAdmin,
+					Verified:     false,
+				},
+			},
+			wantErr:             false,
+			wantVerificationErr: true,
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got, err := tt.s.FromData(tt.args.ctx, tt.args.verify, tt.args.data)
+			if (err != nil) != tt.wantErr {
+				t.Errorf("OpenAIAdmin.FromData() error = %v, wantErr %v", err, tt.wantErr)
+				return
+			}
+			for i := range got {
+				if len(got[i].Raw) == 0 {
+					t.Fatalf("no raw secret present: \n %+v", got[i])
+				}
+				if (got[i].VerificationError() != nil) != tt.wantVerificationErr {
+					t.Fatalf("wantVerificationError = %v, verification error = %v", tt.wantVerificationErr, got[i].VerificationError())
+				}
+			}
+			ignoreOpts := cmpopts.IgnoreFields(detectors.Result{}, "Raw", "verificationError", "AnalysisInfo", "Redacted")
+			ignoreUnexported := cmpopts.IgnoreUnexported(detectors.Result{})
+			if diff := cmp.Diff(got, tt.want, ignoreOpts, ignoreUnexported); diff != "" {
+				t.Errorf("OpenAIAdmin.FromData() %s diff: (-got +want)\n%s", tt.name, diff)
+			}
+		})
+	}
+}
+
+func BenchmarkFromData(benchmark *testing.B) {
+	ctx := context.Background()
+	s := Scanner{}
+	for name, data := range detectors.MustGetBenchmarkData() {
+		benchmark.Run(name, func(b *testing.B) {
+			b.ResetTimer()
+			for n := 0; n < b.N; n++ {
+				_, err := s.FromData(ctx, false, data)
+				if err != nil {
+					b.Fatal(err)
+				}
+			}
+		})
+	}
+}