Sthenos Cyber Division: Offensive and Defensive Security Testing

The Sthenos Cyber Division is the security testing practice of Sthenos Technologies. It attacks a client’s systems the way a real adversary would, through penetration testing, red teaming and AI and LLM red teaming, and then measures whether the client’s defenses would have seen it, through adversary emulation and detection validation. Every engagement starts with signed written authorization, is mapped to NIST SP 800-115, the OWASP testing guides and MITRE ATT&CK, and ends in an evidence-backed report, with a retest available to confirm each fix.

What the Cyber Division does

Most security vendors sell one half of the picture. A penetration test tells you whether an attacker can get in. A monitoring product tells you what your tools logged. Neither tells you the thing a board actually wants to know: when a real attacker does these specific things, would we notice? The Cyber Division runs both halves as one loop, so every engagement ends with a measured answer rather than a list.

Our nine security services

Each service below names what you walk away with and whether the work is run by tooling with a human reviewing it, or performed by hand by a human tester. A scan is never sold as a penetration test, and the report says plainly which one you bought.

Penetration testing

A penetration test puts a qualified attacker against your systems on a fixed, signed scope and hands you confirmed, exploitable risk rather than a list of theoretical issues. The division tests web applications, APIs, external networks and internal networks including Active Directory, and follows NIST SP 800-115 for structure, the OWASP Web Security Testing Guide for web methodology and MITRE ATT&CK to describe every technique used. Tooling assists the tester with enumeration and scanning, but exploitation decisions are made by a person, and denial-of-service testing is never run unless it is separately and explicitly authorized.

Scoping, testing types, black box versus grey box, pricing drivers and what each compliance framework asks for are covered in full on our penetration testing services page.

AI and LLM red teaming

AI red teaming is adversarial testing of an AI or large-language-model application: a chatbot, an agent that takes actions, a retrieval-augmented (RAG) system or an AI feature inside your product. The goal is to find the ways it can be manipulated into leaking data, ignoring its instructions, producing harmful output or taking actions it should not, before a customer or an attacker does. It combines classic application testing with attack techniques that only exist for AI systems.

Breadth comes from two open-source tools stood up and verified in our lab: garak, NVIDIA’s LLM vulnerability scanner, and PyRIT, Microsoft’s AI red-team framework. Depth comes from the tester, who chains attacks and judges real impact. Findings map to the OWASP Top 10 for LLM and GenAI and, where relevant, MITRE ATLAS, with governance framing from the NIST AI Risk Management Framework. Testing runs against a test instance of your system, never a production model you have not authorized.

Adversary emulation and detection validation

A penetration test asks “can we get in”. Adversary emulation asks “when an attacker does these named things, does your detection stack catch them”. We run real-world attacker techniques, one at a time or as a named-adversary chain, against your authorized estate and record for each one whether it was prevented, alerted on, logged without an alert, or missed entirely. It can run collaboratively with your defenders in the room, or as an unannounced exercise with a small trusted group on your side.

1. EmulateRun an ATT&CK-mapped technique with controlled, non-destructive tooling.
2. TelemetryConfirm what it left in your endpoint, network, identity and cloud logs.
3. DetectCheck whether an alert fired, and how good it was.
4. Tune and re-runWrite the missing detection, for example a Sigma rule, and re-run to watch it fire.

You get an ATT&CK coverage heatmap, a detection-gap report with the fix for each gap, tuned detection content validated by re-running the technique, and a prioritized hardening list. Engagements are sized in three tiers: focused validation of a set of techniques you care about, emulation of a named threat actor relevant to your sector using the MITRE Center for Threat-Informed Defense emulation library, or a recurring program that tracks your coverage improving over time.

Cloud, container and supply-chain security

A traditional network test walks straight past a misconfigured cloud account, an over-permissioned identity, a vulnerable container image, a leaked key in source control or a vulnerable dependency deep in the build. The division covers each of them.

A cloud posture review uses read access, not attack, so it generally does not need your provider’s testing authorization; if you add an attack component it becomes a penetration test and follows those rules instead. If you build software with us, the same scanners can run inside your own CI pipeline as pre-merge gates.

Phishing and social-engineering simulation

A controlled phishing campaign against your own staff, run only with written authorization from an accountable executive and sign-off from your HR and legal teams, measures who clicks, who submits credentials and, just as important, who reports it. Email phishing is the default; voice, SMS or physical pretexting happen only if they are explicitly scoped. Results are for improvement, not blame, and are reported in aggregate unless your own policy and consent framework say otherwise.

Continuous monitoring and retesting

A once-a-year test leaves eleven months in which a new subdomain, a changed firewall rule or a regressed fix goes unnoticed. Continuous testing re-runs attack-surface and vulnerability checks on your agreed assets on a schedule, reports what is new, what changed, what was fixed and what came back, and confirms previously reported findings are still closed. Anything that needs a human is escalated to a tester rather than left in a dashboard.

Consented mobile spyware checks

For people who believe their phone may be targeted, the division checks a device owner’s own backup for indicators of mercenary spyware of the Pegasus and Predator class. It uses the Mobile Verification Toolkit, built by Amnesty International’s Security Lab, against public indicator feeds. This is defensive forensics done only with the owner’s consent and on devices they own or the client owns. Sthenos detects this kind of spyware; it never deploys it.

The lab and toolchain behind the work

Every service above runs on tooling the division has installed, version-pinned and proven in its own lab against deliberately vulnerable targets it owns, such as OWASP Juice Shop and DVWA, before it is ever pointed at a client. The stack is drawn from the current open-source frontier rather than one vendor’s suite, so each tool can be swapped when something better appears.

Alongside that toolchain, the division is developing AI-assisted testing agents, still being benchmarked, that run on models hosted on hardware Sthenos owns. Where they are used, they propose and sequence steps, a human approves each exploitation step before it runs, and the AI side of the work does not send your target data to a third-party AI service.

How an engagement runs

1. Scope and authorizationAssets, IP ranges, testing windows, rules of engagement, emergency contacts and a stop procedure are agreed and signed. Nothing touches your systems before this.NIST SP 800-115 planning
2. ReconnaissanceWe map what is reachable: domains, subdomains, live hosts and exposed services, inside the signed scope only.Discovery
3. Scanning and enumerationPorts, services, versions, content and configurations are enumerated and merged into one target map.Discovery
4. Vulnerability analysisCandidate weaknesses are assembled from scanners and manual review, and a human clears the false positives.OWASP WSTG
5. ExploitationControlled exploitation to the depth you authorized, each step approved by a person before it runs.Attack
6. Post-exploitationWhere in scope, privilege escalation and lateral movement to show real business impact, recorded step by step.MITRE ATT&CK
7. ReportingFindings with severity, reproduction steps, evidence and remediation, plus an executive summary and a debrief with your engineers.Reporting
8. RetestThe same tool that proved each finding is re-run against your fix, and every finding is marked fixed, partially fixed or still open.Verification

What lands on your desk

• A findings report: severity, reproduction steps, evidence and remediation for every issue
• An executive summary written for non-technical leadership
• An attack-path narrative for internal and Active Directory tests
• An ATT&CK coverage heatmap for emulation and purple-team work
• Tuned detection rules, validated by re-running the technique
• A CIS-benchmarked posture score and fix roadmap for cloud reviews
• A technical debrief to hand findings to your team
• A retest result for each finding: fixed, partially fixed or open

The rules we work by

Standards every result is measured against

• NIST SP 800-115, Technical Guide to Information Security Testing and Assessment
• OWASP Web Security Testing Guide, the OWASP Top 10 and OWASP ASVS
• OWASP Top 10 for LLM and GenAI and MITRE ATLAS for AI systems
• MITRE ATT&CK for technique mapping and coverage
• CIS Benchmarks and CIS Controls for configuration and continuous vulnerability management
• NIST Cybersecurity Framework and the NIST AI RMF for framing
• The Penetration Testing Execution Standard for engagement structure

If you are building software that has to pass an audit, our guides to FedRAMP-ready development and PCI DSS-ready development cover the build side. Government buyers can find our registrations on the capability statement.

Frequently asked questions

What is the Sthenos Cyber Division?

It is the security testing practice of Sthenos Technologies, a US software and AI company in Tysons, Virginia. The division runs offensive testing, including penetration tests, red teaming and AI and LLM red teaming, together with defensive validation, including adversary emulation and detection engineering, so a client learns both whether an attacker can get in and whether their defenses would notice.

What is the difference between a vulnerability scan and a penetration test?

A vulnerability scan uses automated tools to find known weaknesses and exposed services quickly and broadly, without exploiting anything. A penetration test is performed by a human tester who confirms which weaknesses are actually exploitable, chains them together and finds logic flaws that scanners cannot see. We sell both and every report states plainly which one was performed.

What is AI red teaming?

AI red teaming is adversarial testing of an AI or LLM application, such as a chatbot, agent or RAG system, to find prompt injection, data leakage, jailbreaks and unsafe actions before attackers do. We map every finding to the OWASP Top 10 for LLM and GenAI and, where relevant, to MITRE ATLAS, and test against a test instance of your system.

What is adversary emulation, and how is it different from a penetration test?

A penetration test asks whether an attacker can get in. Adversary emulation runs specific real-world attacker techniques, mapped to MITRE ATT&CK, and measures technique by technique whether your detection stack prevented, alerted on, logged or missed each one. The result is a coverage heatmap and the detection rules to close each gap.

Do you need written authorization before testing?

Yes, always. No scanning, testing, phishing or exploitation begins until an accountable owner of the target signs written authorization naming the scope, the IP ranges or domains, the testing window and the rules of engagement. Phishing additionally needs your HR and legal sign-off.

Do you use AI in your testing, and where does our data go?

Our core toolchain is conventional security tooling run by people. The division is also developing open-source AI-assisted testing agents, still being benchmarked; where they are used, a human tester approves each exploitation step before it runs, and because the models run on hardware Sthenos owns, the AI side of the work does not send your target data to a third-party AI service.

Can you check whether my phone has Pegasus spyware?

Yes, with your consent and on a device you own. We analyze your own device backup with the Mobile Verification Toolkit, built by Amnesty International’s Security Lab, against published indicators for mercenary spyware of the Pegasus and Predator class, and report what was and was not found.

Are your testers certified?

We do not make blanket certification claims. Each proposal names the tester assigned to the engagement and the credentials your requirement calls for, such as OSCP or GPEN, so you can verify the person doing the work rather than a logo on a website.

How much does a security engagement cost?

Every engagement is quoted on its scope: the number of applications, hosts, cloud accounts, AI interfaces or employees in scope, and the depth of testing you authorize. Tell us what you need tested and what is driving the requirement, and we will come back with a written proposal.

Ready to test what an attacker would see? Tell us what is in scope and what is driving the requirement, and the Cyber Division will come back with a proposal that names the tester, the standards and the retest terms. Talk to the Cyber Division or call +1 301-793-3980.