Agent-boundary benchmarks your CISO can reproduce

Swipe to compare vendors

ScenarioBleklineUngovernedLakera GuardKong AI GWOneCLI
B1Secrets in prompt contextPassFailPassPartialPartial
B2Destructive tools/callPassFailN/APartialN/A
B3Session lineage after injectionPassN/AN/AN/AN/A
B4Enforce latency overheadPassN/APartialPartialN/A
B5Credential in tool argsPassFailN/AN/APartial
B6Agent pod egress bypassPassFailN/AN/APartial
B7Time to first governPassN/APartialPartialPartial
B8Audit artifact qualityPassFailPartialPartialPartial
B1
Secrets in prompt context

Can raw secrets reach the model context?

BleklinePass
UngovernedFail
Lakera GuardPass
Kong AI GWPartial
OneCLIPartial
B2
Destructive tools/call

Is tools/call with destructive args blocked at enforce hop?

BleklinePass
UngovernedFail
Lakera GuardN/A
Kong AI GWPartial
OneCLIN/A
B3
Session lineage after injection

After injection, are destructive tools blocked on contaminated session?

BleklinePass
UngovernedN/A
Lakera GuardN/A
Kong AI GWN/A
OneCLIN/A
B4
Enforce latency overhead

What is p50/p95/p99 ms on canonical enforce path?

BleklinePass
UngovernedN/A
Lakera GuardPartial
Kong AI GWPartial
OneCLIN/A
B5
Credential in tool args

Are credentials masked or blocked before tool execution?

BleklinePass
UngovernedFail
Lakera GuardN/A
Kong AI GWN/A
OneCLIPartial
B6
Agent pod egress bypass

Can agent pods reach external HTTP without mandatory hop?

BleklinePass
UngovernedFail
Lakera GuardN/A
Kong AI GWN/A
OneCLIPartial
B7
Time to first govern

Minutes from zero to first enforced interaction?

BleklinePass
UngovernedN/A
Lakera GuardPartial
Kong AI GWPartial
OneCLIPartial
B8
Audit artifact quality

Structured allow/mask/block metadata for SIEM?

BleklinePass
UngovernedFail
Lakera GuardPartial
Kong AI GWPartial
OneCLIPartial

Enforce latency (B4)

p99 on canonical payload · Blekline target <10ms local enforce

Blekline
0.08 ms
Kong AI GW
63 ms
Lakera Guard
306 ms

How we benchmark

Identical payloads across every system. Pass = blocked or masked before execution. Partial = detected but not structurally enforced. Fail = payload reached execution context.

  • 8 scenarios (B1–B8) aligned to NHIM buyer questions
  • Raw JSON artifact with git SHA and version pins
  • Honest N/A where products operate at different layers

Last run: 2026-07-29 · full mode

AI Ingress Control Plane

Build Production Agent Interaction Governance

Agent RUntimes
CLaude COde
Cursor
VSCode
Codex
SDK / REST
Prompt
Tool call
Interaction Governance

Blekline Control Plane

Blekline masks and enforces prompts and tool calls before models and APIs, redacts responses on the return path, with metadata-only audit for security review.
Open core
MCP + API gateway
Metadata-only audit
Ingress
Mask / ENforce / Block
Egress
REsponse / MAsk
Audit
Metadata EXport / SIEM
Masked
Enforced
Audited
Model API'S
Claude
OPenai
Gemini
POlicy
Enoforced
Tools
MCP servers / internal APIs / Webhooks
Developer
VSCode
Terminal
Claude
Cursor
Codex
Prompt
Context
Tool Call
Ingress

Blekline Ingress

Policy enforcement, PII masking, approval queues and audit trails, with mask/block/send options.
Zero retention
Role-Aware
SOC2 Ready
MCP, agent-first
Sanitized
Approved
Auditable
Models
Claude
OPenai
Gemini
Execution
sandboxes
Redacted data
ENvironment
Daytona / Modal / Cloudflare / Vercel

02 - FAQ's

Frequently asked questions

Scoring & methodology

What does Pass vs Partial vs Fail mean?

Pass — blocked or masked before model/tool execution. Partial — detected or HTTP/content-only, not structural MCP enforce at the agent hop. Fail — payload reached execution unchanged.

Why are some matrix cells N/A?

N/A = outside that product’s layer, not a hidden fail. Lakera classifies prompt content (B1), not MCP tools/call (B2/B3). Kong enforces routes/plugins — no session lineage (B3). OneCLI is doc-verified on egress until local proxy is available.

How is audit quality scored in B8?

0–3 scale: 3 = allow/mask/block + findings + requestId; 2 = decision + timestamp; 1 = log/boolean; 0 = no metadata. Blekline scores 3; partial vendors typically 2.
Stack & competitors

Why are Lakera and Kong Partial on some rows?

Lakera flags prompt content — Pass on B1-style injection, N/A on tool enforce. Kong applies route plugins — Partial when the gateway hop exists but tool-arg mask or lineage is out of scope. Partial = detects or routes, not structural enforce at the agent boundary.

How is this different from an LLM firewall or prompt filter?

Firewalls inspect text. B1–B8 score the full agent path — mask, destructive tools/call, lineage, egress (B6), time-to-govern (B7), audit metadata (B8). Complementary, not the same layer.

Does Blekline replace Kong or Okta?

No. Kong = API routes. Okta/Veza = people and access. Blekline = what agents do at runtime. Regulated teams typically need all three. These benchmarks show where each layer stops.
Reproduction & diligence

Can we reproduce these results in our environment?

Yes — gitignored env.benchmark, pnpm benchmark:run, dated JSON with git SHA. CI runs --quick on every build.

Will enforcement slow down our agents?

B4 charts p99. Blekline local enforce targets <10 ms; cloud mask adds RTT. Lakera/Kong include API/gateway hops — compare at the layer each product owns.

Who sees our data when we run a reproduction?

Your keys only — never commit secrets. Cloud mask over TLS; audit defaults to metadata only. Sidecar = in-VPC enforce. EEA-first SaaS; US-primary on enterprise addendum.