Agent shield
TOUGH LOVE SECURITY (marketed at agentshield.win) runs a 13-agent autonomous pentesting pipeline covering recon, attack-surface mapping, parallel vulnerability analysis, conditional exploitation, and reporting, with a senior security engineer reviewing every finding before delivery. It targets web applications and REST/GraphQL APIs, deploys locally across x86, ARM64, Android, macOS, and Linux without Docker or a cloud VM, and is sold in flat-fee tiers (Free to $10K+ Enterprise) plus a $29/mo GitHub Action integration, aimed at SaaS, fintech, healthtech, and medical/dental/behavioral-health practices needing HIPAA-related evaluations. Assessments require signed client authorization and are scoped to a written testing agreement; the vendor carries professional liability insurance capped at the engagement fee and retains assessment data for 90 days before deletion. The vendor also discloses a live AI honeypot used to harden its own classifier against adversarial prompts.
Trust gap
How this is scoredAre scope, spend, and authority enforced while the agent runs?
Scope is limited by signed contract to systems listed in the testing agreement, but the limit is contractual rather than enforced at run time.
How this gap gets closed →Is there a tamper-evident record of what it actually did?
No public evidence found on the page checked: no audit trail, log retention or customer-reviewable record is described.
How this gap gets closed →Are compliance obligations attached per engagement?
Publishes a safety/authorisation policy and markets HIPAA-related assessments, but publishes no certification of its own.
How this gap gets closed →Does payment depend on a verified result?
A 30-day money-back guarantee is tied to a 10-day delivery deadline, so a fee is refundable on a delivery outcome, not on a verified security result.
How this gap gets closed →Can the deployment be insured, and is the customer indemnified?
States it carries professional liability insurance and caps liability at the engagement fee; no carrier is named, no cover limit is published, and no customer indemnity is offered.
How this gap gets closed →Assurance
- Insurance available
- Vendor carries professional liability insurance
“We carry professional liability insurance.”
agentshield.win · checked Aug 2, 2026 - Insurance carriers
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Coverage limits
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Indemnification
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Liability cap
- Liability contractually capped at the engagement fee paid
“Our liability is limited to the engagement fee paid.”
agentshield.win · checked Aug 2, 2026 - Audit trail
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Tamper-evident log
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Explainability
- Every finding comes with a working proof-of-concept as evidence of why it was flagged
“Every finding includes a working proof-of-concept.”
agentshield.win · checked Aug 2, 2026 - Runtime governance
- Testing confined by contract to systems explicitly listed in the signed scope document
“Testing is conducted exclusively within the boundaries defined in the testing agreement. Any systems, endpoints, or networks not explicitly listed in the scope document will not be tested.”
agentshield.win · checked Aug 2, 2026 - Permission scopes
- Client supplies multi-role test credentials, source code access, API documentation and architecture diagrams
“test credentials (multiple role levels), source code access (private repo or zip), API documentation, and architecture diagrams.”
agentshield.win · checked Aug 2, 2026 - Compliance certifications
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Regulatory alignment
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Outcome-based pricing
- 10-day delivery with a 30-day money-back guarantee if missed
“10-day delivery. 30-day money-back if we miss.”
agentshield.win · checked Aug 2, 2026 - Settlement mechanism
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Dispute process
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- SLA terms
- Enterprise tier includes SLA-backed response times
“SLA-backed response times”
agentshield.win · checked Aug 2, 2026 - Evaluation coverage
- Self-administered benchmark testing against OWASP DVWA/Mutillidae (not stated as independent)
“Tested against OWASP DVWA (25 known vulns) and Mutillidae (30 known vulns) across 4 rounds of iterative training.”
agentshield.win · checked Aug 2, 2026
Agency
- Autonomy level
- AI agents perform testing/exploitation autonomously; a human validates each finding before delivery (not full unsupervised autonomy)
“The AI agents do the heavy lifting — recon, data flow analysis, payload generation, exploitation — but a human validates each finding, removes false positives, and writes remediation guidance.”
agentshield.win · checked Aug 2, 2026 - Human oversight
- Senior security engineer reviews and approves every finding before delivery
“Every finding is reviewed by a senior security engineer before delivery.”
agentshield.win · checked Aug 2, 2026 - Goal complexity
- Multi-step conditional pipeline: recon, surface mapping, vulnerability analysis, conditional exploitation, reporting
“Each phase feeds the next. Exploitation is conditional — we only attempt it when analysis confirms a real vulnerability.”
agentshield.win · checked Aug 2, 2026 - Action space
- Non-destructive PoC actions only: read-only data exfiltration, session capture, privilege verification (no DoS, deletion, or production changes)
“Exploitation attempts are limited to proof-of-concept demonstrations — read-only data exfiltration, session capture, or privilege verification.”
agentshield.win · checked Aug 2, 2026 - Operating environment
- Operates against web applications and REST/GraphQL APIs
“Web applications and APIs (REST, GraphQL).”
agentshield.win · checked Aug 2, 2026 - Initiative
- Acts only after client scopes the target and signs authorization; will not test without authorization
“All assessments require explicit written authorization from the target owner. We do not test systems without proper authorization under any circumstances.”
agentshield.win · checked Aug 2, 2026
Safety
- Safety evaluations
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Red teaming
- Runs a live AI honeypot capturing jailbreak attempts, publishes results to a public ledger, and reports its classifier blocks 100% of adversarial prompts in held-out testing
“Live AI honeypot · captures jailbreak attempts → hardens TUFFY classifier within 30 min · public ledger at /constitutional · TUFFY blocks 100% of adversarial prompts on our held-out fair-test”
agentshield.win · checked Aug 2, 2026 - Safety policy
- Published policy requiring signed authorization/testing agreement before any assessment; testing confined to defined scope
“Authorization Required: All security assessments require explicit written authorization from the system owner. TOUGH LOVE SECURITY will not conduct testing without a signed testing agreement that defines scope, authorized targets, and testing boundaries.”
agentshield.win · checked Aug 2, 2026 - Usage restrictions
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Model or system card
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Incident reporting
- Has a responsible-disclosure practice for third-party vulnerabilities discovered during assessments
“If during an authorized assessment we discover vulnerabilities affecting third-party systems or data belonging to parties other than the client, we will notify the client immediately and follow responsible disclosure practices.”
agentshield.win · checked Aug 2, 2026 - Third-party evaluations
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Data handling
- Assessment data retained 90 days post-delivery for re-test purposes, then permanently deleted; early deletion available on request
“Assessment data is retained for 90 days after report delivery for re-test purposes, then permanently deleted. You may request early deletion at any time.”
agentshield.win · checked Aug 2, 2026
Practicality
- Pricing model
- Charged per engagement/per target (flat-fee tiers), plus a $29/mo subscription tier for GitHub Action scans
“Full Pentest $2,500 per engagement”
agentshield.win · checked Aug 2, 2026 - Price point
- HIPAA Bundle priced at $3,500 flat; tiers range from Free to $10K+ Enterprise
“First Pass Your HIPAA Audit in 30 Days — $3,500 Flat”
agentshield.win · checked Aug 2, 2026 - Availability
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Deployment options
- Runs locally across x86, ARM64, Android, macOS, Linux without Docker or a cloud VM; scans execute locally
“Runs on x86, ARM64, Android, macOS, Linux. No Docker required. No cloud VM.”
agentshield.win · checked Aug 2, 2026 - Integrations
- Integrates with GitHub Actions (PR-triggered scans) and Slack (alerts)
“GitHub Action tier — $29/mo (20 scans/mo, PR-triggered, Slack alerts).”
agentshield.win · checked Aug 2, 2026 - Supported regions
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Support model
- Quarterly tier includes a dedicated Slack channel; Enterprise includes a dedicated security advisor
“Dedicated Slack channel”
agentshield.win · checked Aug 2, 2026
Foundation models
- Base models
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Model provider
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Model swappable
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Open weights
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Fine-tuning
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Context window
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
Ecosystem
- Protocols supported
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Tool use
- Uses external tools including nmap, subfinder, whatweb, and code-review analysis for reconnaissance
“nmap subfinder whatweb code-review”
agentshield.win · checked Aug 2, 2026 - Multi-agent
- Multiple specialist AI agents run in parallel (13 agents total; 5 run simultaneously during vulnerability analysis)
“Five specialist agents run simultaneously. Each performs source-to-sink taint analysis with code-backed evidence.”
agentshield.win · checked Aug 2, 2026 - API access
- GitHub Action integration provides PR-triggered automated scanning
“GitHub Action tier — $29/mo (20 scans/mo, PR-triggered, Slack alerts).”
agentshield.win · checked Aug 2, 2026 - Open source
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Marketplace presence
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
Impact
- User base
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this
- Deployment scale
- Founder describes it as a young/early-stage company, no large-scale deployment figures given
“This is a young company. I'm transparent about that.”
agentshield.win · checked Aug 2, 2026 - Target sectors
- Sold into SaaS, fintech, healthtech, plus dental/medical/behavioral-health practices
“SaaS companies, fintech, healthtech, and anyone whose customers or regulators expect regular security validation.”
agentshield.win · checked Aug 2, 2026 - High-risk domains
- Targets healthcare (medical, dental, behavioral-health) — a high-risk/regulated domain (HIPAA/OCR)
“Built specifically for small medical, dental, and behavioral-health practices preparing for OCR.”
agentshield.win · checked Aug 2, 2026 - Documented incidents
- no public informationagentshield.win · checked Aug 2, 2026 · source did not state this