End to end. On its own.
Full security engagements without human intervention during the run. A human reviews the output.
Autonomous AI penetration testing
The whole job.
On its own.
Full security engagements, from planning to verified findings. No human intervention during the run. A human reviews the output.
Request pilot access
L4 today. Building toward L5.
Private development · Pilot opening soon
Private development · Pilot opening soon
Autonomy / current position
04 — 05
CurrentEnd-to-end engagements
TrajectoryL5 — not yet achieved
Saifullah’s public marketing framework
01 / The autonomy scale
L4 today.
Building toward L5.
The distinction is who carries the engagement — and when a human steps in.
- L0 Manual A human runs every tool and interprets every result.
- L1 Assisted Scanners flag candidates. A human verifies each one.
- L2 Partial automation Scripted checks. A human drives the engagement.
- L3 Supervised autonomy AI executes tasks. A human approves each step.
- Saifullah today L4 High autonomy The full engagement. A human reviews the output.
- Not achieved L5 Full autonomy Novel situations. No human involvement.
L0 / Manual
Every action is human-led.
A human runs each tool, decides what comes next and interprets every result.
Human involvement throughout.
L1 / Assisted
Candidates still need an operator.
Scanners flag potential issues. A human verifies each candidate and carries the engagement forward.
A human verifies every candidate.
L2 / Partial automation
Checks run. A human drives.
Scripted checks run in sequence; a human drives every engagement and interprets the results.
A human directs the engagement.
L3 / Supervised autonomy
Execution, with a human gate.
AI executes bounded tasks while a human approves each step. The engagement still depends on those approvals.
Human approval at each step.
L4 / High autonomy — Saifullah today
No handoffs during the run.
Planning, reconnaissance, discovery, chaining, verification and reporting — without human intervention during the engagement.
A human reviews the output.
L5 / Full autonomy — not achieved
The engineering north star.
Handling novel, unanticipated situations and adapting with no human involvement. This is a goal, not an achieved capability.
Not achieved. Being built toward.
Saifullah’s public marketing framework — not an industry standard or an internal architecture diagram.
Illustrative visualizer · Select a level02 / Anatomy of an engagement
From plan to proof.
Illustrative sequence · Sample states
One autonomous engagement
No human touchpoints during the run.
Reduced motion: static steps. Automatic playback is off.
-
Plan the security engagement.
-
Explore the application surface.
-
Identify vulnerability candidates.
-
Connect issues into attack paths.
-
Reproduce findings and their impact.
-
Deliver evidence-backed findings.
Illustrative frame 06 / 06 — Report
Human review follows the engagement
Evidence-backed findings are prepared for human review. The autonomous run is complete.
The finding, not the noise
Nothing reported
without proof.
A candidate is not a finding. Every reported vulnerability is backed by reproduced evidence — not just an alert.
Reproduced evidence
Multi-step attack paths
Demonstrated impact
Illustrative finding / sample states
Reported
Evidence attached. Ready for human review.
Sample evidence categories — not actual artifacts
Proof of conceptShown in sample
Response logsShown in sample
Demonstrated impactShown in sample
Only evidence-verified findings enter the report. A human reviews the output.
Reduced motion: static steps. Automatic playback is off.
Illustrative sample only. No real target, vulnerability or live system data.
Sequence and finding shown for explanation only. Neither is connected to a live Saifullah engagement.
03 / What the system carries
Autonomy needs substance.
Not a conversational layer over a scanner. A system that does the engagement.
Evidence, not alert volume.
Every reported vulnerability is backed by reproduced evidence: proof of concept, response logs and demonstrated impact.
Attack paths, not isolated issues.
Multi-step exploit chaining connects individual issues into the paths an attacker could take.
Changes happen. The work continues.
Adapts to application changes and new deployments without manual re-engagement.
Beyond a single surface.
Web applications, APIs, desktop applications and mobile surfaces — operated the way a human operator would.
Across surfaces
Web applications
APIs
Desktop
Mobile
04 / The landscape
Not a scanner.
Not a wrapper.
The difference is not the interface. It is who does the work — from the first plan to the verified report.
Approaches, not a vendor leaderboard.Automated scanners
Candidate alerts still need to become verified findings.
Manual pentests
Human expertise, delivered as a point-in-time engagement.
AI wrappers
A conversational interface is not, by itself, autonomous execution.
Saifullah L4
The system carries the full engagement. A human reviews the evidenced output.
L4 within Saifullah’s public framework. L5 remains the goal.
Private development
Pilot opening soon.
A limited program for select companies. The waitlist is open.
Built by Saif
One founder. Full-time.
Saif is building Saifullah from the ground up. Technical founder. Deep in the security work.
The next step
Request
pilot access.
For companies looking beyond another scan. Join the waitlist, or contact the founder directly.
contact@saifullahsec.com