Cobalt delivers scheduled, human-run pentests through a marketplace of testers. Planck Operator is a steerable, autonomous agent that runs continuously between engagements, returns exploit-proven findings, and offers a self-serve start with a free Recon tier so you can begin without a scoping call.
Cobalt is a Pentest-as-a-Service marketplace: you buy time-boxed engagements delivered by human testers. Planck Operator is a continuous agent you can steer. The two answer different questions, and the difference shows up in cadence, pricing, and how findings arrive.
| Planck Operator | Cobalt | |
|---|---|---|
| Delivery model | Steerable autonomous agent, human-on-the-loop | Marketplace of human pentesters |
| Cadence | Continuous re-testing | Time-boxed, scheduled engagements |
| Getting started | Self-serve, free Recon tier | 8-hour credit model, generally a scoping call to convert to scope |
| Pricing | Self-serve on-ramp, free Recon tier, plans scale by domain & depth | Engagement median ~$30k (third-party estimate) |
| Findings | Exploit-proven: request/response, repro, CVSS v3.1 | Human report per engagement |
| Compliance letter | Human validation on demand | Signed letter per engagement |
Cobalt has also launched Cobalt Autonomous, an AI pentest offering priced around ~$3,500 (third-party estimate). Planck Operator differs in being steerable and human-on-the-loop with a self-serve on-ramp rather than a scoping-call credit model.
Cobalt built a mature marketplace of vetted human pentesters, and that is a real strength. When you need a scheduled, deep engagement with a named human team and a signed compliance letter at the end, a PTaaS marketplace is a proven way to get one.
For teams whose primary need is a point-in-time assessment on a fixed calendar, delivered by people, Cobalt is a credible choice. We do not position Planck Operator as a replacement for that work. We position it as the continuous, steerable layer that runs alongside it.
The gap a scheduled pentest leaves is time. Your attack surface changes every deployment; an engagement every few months cannot see the exposure shipped last Tuesday. Planck Operator closes that gap.
For continuous coverage, yes. Planck Operator runs autonomously and continuously between the deep, scheduled engagements a human marketplace like Cobalt delivers. Many teams keep a periodic human pentest for a signed compliance letter and run Planck Operator to hold the line every day in between.
Planck Operator offers a self-serve on-ramp with a free Recon discovery tier, then paid plans that scale by verified domain and depth. Cobalt's human PTaaS engagements have a median around $30,000 per engagement (third-party estimate) and generally begin with a scoping call to convert credits into scope.
As of 2026 Cobalt has launched Cobalt Autonomous, an AI pentest offering priced around $3,500 (third-party estimate). Planck Operator differs in being steerable and human-on-the-loop, with exploit-proven continuous findings and a self-serve on-ramp rather than a scoping-call credit model.
Planck Operator sits between fully-autonomous and human-validated testing. It runs on its own for breadth, but a human can stay on the loop, directing it at specific business logic, edge cases, and priorities. You keep autonomy's constancy without giving up the ability to steer.
Yes. Every Planck Operator finding is exploit-proven and ships with the request and response, reproduction steps, and a CVSS v3.1 severity. It is proof, not probability, so your engineers can confirm and fix without re-triage.
Keep your deep engagements. Add an agent that holds the line between them, with proof attached to every finding.