PenTrust / Control Tower
Bobby Venkatesh
Administrator
BV
Sign out
Engagement · shop.acme.com
Control Tower Everything the engine is doing right now, read against the standards — one command deck.
+ New Assessment
7 2 binary · 5 live-target
Findings confirmed 34 ▲ 6 today
Patterns learned +128 corpus growth, 24h
Hallucination rate 0.4% ▼ on proving ground
Live coverage NIST 800-115 · ATT&CK
Checklist floor (MASVS/ASVS) 100%
High-severity leads chased 82%
Technique stream
T1190 Exploit public-facing app — login endpoint now
T1046 Network service scanning — 443/8443 12s
T1592 Gather host information — whatweb 48s
T1110 Brute force — throttled, confirm-only 1m
T1213 Data from information repositories 2m
Agent lifecycle extract-then-incinerate
Spawn →
Run →
Extract →
Incinerate
Findings survive; raw target material dies with the agent. Termination is Lens-enforced on a deadline — never self-trust.
Autonomy posture precision buys autonomy
AUTONOMOUS Discovery & validation 98.1% ✓
SUPERVISED Reversible auto-apply earning
HUMAN ONLY Irreversible changes permanent
Step 1 of 1 · set the envelope
New Assessment Set scope, prove ownership, and define the risk envelope once. Inside it, the engine runs the rest.
Target
Target type
Live portal Binary / APK
Aggression dial
Confirm-only Post-exploit Full-chain
Above confirm-only requires explicit human approval at each gate.
Reversibility line — auto-apply ceiling
Advise only Reversible All
Auth, schema & data changes stay human-only regardless of this setting.
Verify ownership & launch
Domain ownership permanent gate · legal
✓ DNS TXT record — pentrust-verify=9f3a… found
✓ Scope allowlist locked at the container network layer
✓ Signed authorization record created for this engagement
• Launch — waiting on confirmation
An agent physically cannot reach a host outside this allowlist. Any attempt is killed instantly and sealed.
Live Job The engine working at machine speed; the Lens reading every log line against the standards.
Pause job
Chunks done 18 / 24
Findings 41
Verified 34 7 dropped by Shadow
Elapsed 22m
Technique execution feed
T1190 SQLi probe → benign version read now
T1059 Command injection test — no callback 9s
T1552 Unsecured credentials — .env exposed 31s
T1071 App-layer protocol — TLS fingerprint 1m
T1505 Server software component scan 2m
T1083 File & directory discovery 3m
Agent #7 live-target
Spawn ✓
Run ✓
Extract findings…
Incinerate
⚠
Approval required — aggression escalation
Shadow requests post-exploitation on the exposed .env . This crosses an action-budget gate.
Deny Approve once
34 confirmed · 7 indicated · 2 novel
Findings Explorer Honest grading, every time. Confirmed means proven safely; indicated means real but not worth breaking production to detonate.
ATT&CK: any
Status: any
Novel only
SQL injection on /api/login
T1190 · CVSS 9.1 · leaks customer email table
CRITICAL ✓ CONFIRMED
Exposed .env with DB credentials
T1552 · CVSS 8.6 · plaintext secrets
CRITICAL ✓ CONFIRMED
Reflected XSS on search param
T1059 · CVSS 7.2 · marker callback fired
HIGH ✓ CONFIRMED
SSRF via webhook URL field
T1190 · CVSS 7.5 · safe internal string pulled
HIGH ◴ INDICATED
Unusual auth-bypass chain
nearest tactic: initial-access · escalated to research
✦ NOVEL ◴ INDICATED
Outdated TLS cipher suite
T1071 · CVSS 5.3 · corpus pattern match
MEDIUM ✓ CONFIRMED
Crown jewel · the solution is the product
Remediation Package Not just the problem — the exact, standard-cited fix, with an honest note on what it should not break.
Finding
SQL injection on /api/login
T1190 · CVSS 9.1
CRITICAL ✓ CONFIRMED ↩ REVERSIBLE FIX
Data-leak path
Unparameterised query lets an attacker read the full customers table — names, emails, hashes.
Blast radius
Entire customer PII store. Credential-stuffing risk downstream.
The fix ordered · safe to apply
1 Replace string-built query with a parameterised statement (prepared bind)
2 Add input validation on the username field at the API boundary
3 Rotate any credentials exposed during the window
Cited, not asserted says who?
MITRE Mitigation M1013 — Application Developer Guidance
OWASP ASVS 5.3.4 — Parameterised queries
Non-breaking analysis
Should not affect valid logins. Recommended: validate in staging first.
Advise only Apply in staging
The deliverable is the path, not the list
Attack Graph Findings chained into adversary progression — entry technique through to the crown jewels.
Entry T1190 SQLi crit recon .env creds tls PII crown
The hero report artifact
ATT&CK Heatmap Which adversary techniques were attempted, and which landed — keyed to technique IDs.
The driver · contained, never the boundary
AI Shadow Autonomy earned by measured precision — never granted by decision. Two budgets, two kinds of walls.
Drive 82% of loop steps autonomous
Verify 99.6% precision on proving ground
Persist 3 surface changes re-driven
Gate taxonomy precision buys autonomy
🔓 Next-move & triage
Trust wall — autonomy climbs with precision
🔓 Confirm-only validation
Trust wall — on an already-authorized target
🔓 Reversible remediation apply
Trust wall — within the owner envelope
🔒 Authorization & scope-lock
Legal wall — never automated
PERMANENT
🔒 Irreversible change (auth/schema/data)
Liability wall — human only, any confidence
PERMANENT
Sharper today than yesterday
Living Corpus One shared engine that learns from every engagement. Patterns travel — raw data never leaves the perimeter.
Patterns held 48.2k
Learned today +128
Corpus version v3.42 signed · gated
Promoted (de-id'd) 36 pattern, not data
Pattern lifecycle fail-closed de-identification
✓ Confirmed finding → generalised (loci stripped)
✓ De-identified → promotion queue (opt-in)
✓ Central proving-ground gate → signed bundle
• Shipped to every appliance tonight
Recent learnings internal KB
Auth-bypass chain on JWT refresh
novel · nearest T1212
✦ NOVEL Prototype pollution via query merge
new-to-engine · T1059
LEARNED Cache-key confusion in CDN layer
new-to-engine · T1557
LEARNED
Darkness is armour, never cover
Audit Trail Every action, remediation and autonomous decision — hash-chained and tamper-evident.
172 Shadow decision · confirm-only validation · earned basis pg-scorecard #4471
sha256: a9f3c2…7e10
171 Remediation issued · SQLi /api/login · M1013, ASVS 5.3.4
sha256: 3b8e11…c2a4
170 Finding confirmed · T1190 · evidence sealed
sha256: 77c0ab…19de
169 Agent incinerated · #6 · verified teardown
sha256: 1f44d9…8b02
The product's value surface
Report Viewer One layered report — exec narrative, technical detail, remediation, delta and the sealed methodology trail.
Export
Confidential · shop.acme.com
Security Assessment — October 2026
Supervised-autonomous engagement · 34 confirmed findings · methodology sealed (NIST SP 800-115)
PenTrust walked the full attack chain against the Acme storefront and confirmed a critical path from an unauthenticated SQL injection on the login endpoint through to the full customer PII store. Each finding below carries its ATT&CK technique, CVSS vector, validation status, and a standard-cited remediation.
Critical 4
High 11
Medium 19
All four critical findings include reversible, staged remediation packages. Re-test delta against the September run shows 9 previously-open issues now closed.
Administration · Directory
User Administration Create and manage accounts, reset credentials, block access, and assign role-based permissions across the PenTrust portal.
+ Create user
Total users 6
Active 5
Blocked 1
Roles defined 5
Roles & access role-based access control
Capability Admin Operator Analyst Auditor Viewer
Run & manage assessments ✓ ✓ — — — Approve aggression / apply fixes ✓ ✓ — — — View findings & remediation ✓ ✓ ✓ ✓ ✓ Tune AI Shadow autonomy ✓ — — — — Read audit trail ✓ ✓ ✓ ✓ — Export reports ✓ ✓ ✓ ✓ — Manage users & roles ✓ — — — —
