cybersecurity, identity & threat intelligence·run 49 · Apr 2026

Promptbastion

A CI-oriented security test service for authorized AI endpoints and MCP servers that binds an attack suite to an exact release and environment, records observable behavior and coverage, and routes findings, remediation, exceptions, retest, expiry, and correction.

Genesis score7.05/10

Make Promptbastion real.0/500

500 more votes and Promptbastion is authorized for build.

0%500 to authorize

Backing is the vote. When an idea crosses 500, we pull it into the build pipeline and ship it for real — the votes decide what gets built next, not an editor.

The opportunity

1+Confirmed direct AI security scanners

1Confirmed isolated test substrates

0Reviewed scanners naming MCP tests

The case

The research confirms a direct pre-deploy AI scanner with CI integration, additional generic scanners, an isolated execution service, and a free signing transparency interface. It reports that reviewed competitors do not mention MCP-specific tests. No structural incumbent copying cost is evidenced.

Promptbastion cannot prove an endpoint or server secure, safe, non-malicious, compliant, or free from prompt injection, data leakage, jailbreaks, permission abuse, vulnerabilities, or future regressions. Test results depend on authorization, target, release, environment, model and provider versions, policies, tools, permissions, data, fixtures, randomness, repetition, evaluator, and time. A named taxonomy is coverage guidance, not certification.

Tests must never target systems without owner and provider authority, use real secrets or customer data by default, escape the approved environment, or execute consequential tools. Findings are observed reproducible cases with uncertainty. Signed bundles establish integrity and signer attribution, not truth, completeness, acquisition readiness, or due-diligence acceptance.

Who pays — and why

Solo founders and small engineering teams shipping AI endpoints or MCP servers who need a repeatable authorized pre-deploy test gate.

What it unlocks

A target authorization with owner, provider and hosting rights, endpoint or artifact, environment, release digest, models, tools, permissions, data classes, network destinations, excluded surfaces, test window, approver, expiry, and revocation

A versioned suite with taxonomy reference, attack case, preconditions, synthetic fixture, expected safe property, tool and data limits, repetition, evaluator, severity method, timeout, stop condition, and known coverage

A result separating deterministic observation, model or judge observation, human review, request and response evidence, tool behavior, data exposure, reproduction, uncertainty, false-positive disposition, affected release, and no permanent PASS

A gate record with required tests, findings, owner, remediation, exception or waiver, expiry, retest, release decision, downstream deployment acknowledgment, readback, rollback, signature, correction, and supersession

How Genesis scored it

7.05across seven criteria

Temporal window

Rapid AI and tool-server deployment makes pre-release testing timely.

Buyer persona

Small teams shipping AI endpoints are clear.

Convergence

Several scanners and security primitives support the category.

Why it scored well

A direct competitor validates demand, the MCP coverage gap is specific, the small-team buyer is clear, and release-bound suites, reproducible cases, CI gates, remediation, retest, and expiry form a buildable wedge.

What's holding it back

The market already has scanners, test coverage is incomplete, dynamic targets are nondeterministic, authorization and sandboxing matter, false confidence is dangerous, and incumbents can add MCP cases.

Signals detected3 sources crossed

SignalGenesis research

Direction briefpromptbastion.md

promptbastion.md

Want this pointed at your vertical?Point Genesis at your own market and constraints — it invents adjacent, fork-ready ideas, private to you before they hit the public feed.

Discussion

No comments yet — be the first to weigh in.