CyberSecAI · Machine-Adversary Resistance

MARS

A rating of how well a system resists an autonomous AI adversary. Every AI-security framework tells you what to do. MARS measures whether what you built actually holds — and makes you prove it.

Paper · DOI 10.5281/zenodo.24341963 IETF · draft-sharif-mars-rating License · CC BY 4.0 Status · proposed, open

Offered as a starting point for a standard, not a ratified standard. Companion to the AgentPass ABCDE Model.

AI attacks are the new norm. The question is no longer whether you are compliant — it is whether you hold, and whether you can prove it.

Why MARS exists

The attacker is no longer human-slow

Security was built for a human adversary: slow, serial, expensive to scale, forgetful between attempts. The autonomous AI adversary is the inverse. It enumerates an attack surface exhaustively at machine speed. It reasons from artefacts — a patch, a diff, an error message — straight to a working exploit. It coordinates as a swarm. It learns across attempts. And it moves laterally with valid-looking credentials.

This is not a forecast. By late 2026 each of these capabilities had been observed operationally. The field has excellent guidance — catalogues of threats and controls — but all of it answers one question: "what should I do?" None of it answers the question that now decides whether a system survives: "if an autonomous AI attacks this, does it hold, and how well?"

Compliance is the input. Resistance is the outcome. A system can satisfy every control catalogue and still be trivially enumerated, re-exploited and swarmed. MARS is the missing outcome rating.

Observed in 2026 — the reason this matters now

16 Nov 2026

Autonomous patch-to-exploit

Google Threat Intelligence Group found a framework on an exposed server that diffs a published patch, infers the fixed weakness, writes an exploit and runs it — no human in the loop. The act of shipping a patch can now hand the attacker the exploit.

13 Nov 2026

Agentic intrusion framework

An FBI FLASH advisory described an autonomous penetration framework in criminal use — some intrusions from first contact to privileged access in under three hours.

2026

Attack swarms

Autonomous pentest agents built on swarms of models reached the top of a major public vulnerability-disclosure leaderboard. Dozens of open-source offensive agents now exist.

The threat model

Five moves of an autonomous AI attacker

MARS is defined against a specific adversary: an AI that operates at machine speed, with no human in the decision loop, and improves across attempts. Resistance to a slower human attacker does not earn a MARS rating. The adversary has five moves — and each of the five MARS axes exists to deny exactly one of them.

Move 1

Learn

Enumerate and map the entire attack surface exhaustively, at machine speed, including authenticated and east-west surface a scanner never reached.

Denied by: axis L

Move 2

Weaponize

Turn an artefact — a published patch, a diff, an error message — into a working exploit automatically. N-day to exploit in minutes.

Denied by: axis R

Move 3

Swarm

Run many coordinated agents in parallel, dividing the surface and sharing what each one learns.

Denied by: axis S

Move 4

Adapt

Model the defence, route around it, and improve across repeated attempts until something works.

Denied by: axis O

Move 5

Spread

Move laterally with valid-looking credentials, blend in, dwell, and pivot deeper into the estate.

Denied by: axis P

The rating is assessed at a stated access tier: Tier 0 (unauthenticated), Tier 1 (the attacker holds a valid credential or session — the demanding, default tier), or Tier 2 (a trusted supply-chain position). A rating must say which tier it was earned against, because resistance that holds against an outsider may not hold against a valid session.
What MARS is

Five things an AI attacker must be stopped from doing

MARS rates any defended surface — a server, a cloud workload, an API, a whole perimeter — because the attacker is the AI, even when the thing being defended is not. A system is resistant to the degree the adversary cannot:

L

Learn it

The surface cannot be enumerated or mapped, even after authentication.

Stops: machine-speed reconnaissance.
Built with: moving-target defence, per-identity surface mutation, zero-trust perimeters.

R

Reuse against it

A patch, diff, recon result or captured request yields nothing reusable.

Stops: patch-to-exploit, replay.
Built with: non-replayable requests, silent patching, implementation diversity, runtime shielding.

S

Scale on it

A coordinated swarm is absorbed, not amplified; one detection protects the whole fleet.

Stops: parallel multi-agent pressure.
Built with: fleet detection-and-response, instant identity quarantine.

O

Outpace it

The defence changes faster than it can be learned, and responds in-request, not as an alert.

Stops: defence-learning and adaptation.
Built with: non-stationarity, in-request reflex.

P

Spread through it

One compromised node is contained in seconds, so lateral movement is prevented.

Stops: pivoting with valid credentials.
Built with: per-service identity, micro-segmentation, instant revocation.

and it must be proven

Above level 2, a rating is earned by measured, tamper-evident evidence — never self-declared. Claiming resistance without proof is "MARS-washing," and the rating rejects it.

The scale: 0 to 5 stars

Higher is more resistant. The overall rating is gated by the weakest axis — one weak axis caps the whole score, exactly like a crash rating.

0
Exposed

Static, signature-based. A single agent defeats it.

1
Hardened

WAF/EDR/MFA. Stops humans, not AI.

2
Reactive

Detects agentic patterns and alerts.

3
Moving

Unmappable and non-replayable (L+R).

4
Reflexive

In-request response + fleet containment.

5
Provable

All axes + continuous measured evidence.

Where it sits in your estate

MARS across the network: strongest at the exposed edge and the critical core

An autonomous adversary meets your internet-facing servers first and works inward. MARS sets a required rating per zone, highest at the two highest-risk zones: a provable internet-facing edge it cannot map or reuse, and a provable core around what matters most. The sheltered middle steps down.

AI adversary autonomous swarm Internet — untrusted attacks ← MARS-protected perimeter Internet-facing public APIs, web, auth MARS 5 Provable · the exposed front line DMZ shared / partner services MARS 4 Reflexive Internal east-west workloads MARS 3 Moving Crown jewels data, secrets, prod control MARS 5 Provable · continuous evidence

Illustrative. Each zone carries the MARS minimum its exposure and value justify; a zone's rating is gated by its weakest external component.

How it is measured

A fixed adversary and fixed metrics — so a rating means the same everywhere

Most scoring schemes are vague about the adversary, so their numbers are not comparable. MARS fixes both the adversary and the measurement. This is the step that turns a rating into something more than an opinion.

The MARS Reference Adversary (MRA)

A versioned, tool-agnostic capability profile — the minimum an attacker must be able to do so that any conforming test rig gives comparable results: full autonomy at machine speed, exhaustive enumeration, automated patch-to-exploit, coordinated swarming, learning across attempts, and lateral movement. Named tools (ARTEX-class, swarm pentesters) are examples that meet the floor, not the definition. The profile is fluid and versioned, revised as attackers advance.

The evidence rule (anti-"MARS-washing")

Above level 2, a rating must be earned by a reproducible result against an MRA-conforming adversary, recording surface mapped (target: zero), captured-exploit reuse (target: zero), patch-to-exploit yield, containment time and lateral hops. Results are written to a tamper-evident, independently verifiable record. Independent verification is what separates a rating from marketing.

Measurable, falsifiable criteria

Each axis scores 0 (absent), 1 (partial) or 2 (full), against numbers, not judgement — for example L is 2 only when the adversary recovers zero live endpoints, R is 2 only when reuse and operated-system patch-to-exploit yield are both zero. A rating must be able to fail.

Stated with its conditions

Because the adversary advances, a rating is time-bound and tier-bound. It is always written with its conditions, for example MARS 4, MRA-1, Tier 1, as of 2026-11-18. A bare number without them is incomplete.

Proof, not slideware

The first measured MARS ratings

MARS is not only a concept. Using the MRA-1 profile we measured two systems for real — a static service and a moving-target service — and recorded every result to a hash-chained evidence ledger that re-verifies. Nothing below is asserted; every figure is observed.

Axis / metricStatic baselineMoving target
L · surface recovered by enumeration1.00 mapped0.00 nothing
R · captured-request replay success1.00 reusable0.00 dead
O · non-stationary + in-request rejectno staticyes moving
S · swarm containmentnot assessednot assessed
P · lateral spreadnot assessednot assessed
RatingMARS 0MARS 3

Honest scope: L, R and O are measured on a single node. S and P require a multi-node fleet harness and are marked not assessed, so this run cannot certify above level 3. Full expression: MARS 3, MRA-1, Tier 0, as of 2026-11-18. Evidence ledger: 8 entries, hash-chained, verified intact and re-runnable.

Reference harness — run it yourself Evidence ledger (hash-chained) Measured report Assessment (JSON) MRA-1 profile

Reproduce it: unzip the harness, run bash run.sh, then verify the ledger with node verify.js. Every figure above regenerates.

How an organisation uses it

A rating you can demand, architect to, and audit

Resistance is a cost, and five-star resistance everywhere is neither necessary nor defensible to a board. MARS is applied the way any mature security control is applied — as a risk-based minimum per zone. Each tier carries the rating its exposure and value justify, and no more: investment is concentrated where an autonomous adversary actually reaches, and the unnecessary cost of over-hardening low-risk internal surface is removed.

Right-sized by risk: a MARS minimum per security zone
Exposure falls as you move inward; value peaks at the core. The required rating tracks both — so spend follows risk, not habit.
Internet-facing edge Public APIs, web apps, authentication — where the swarm hits first MARS 5 Provable DMZ / shared services Partner and edge services behind the perimeter min. 4 Reflexive Internal services East-west workloads not exposed to the internet min. 3 Moving Crown jewels Data stores, secrets, production control — highest value MARS 5 Provable
Illustrative tiering. A zone's rating is gated by its weakest external component — one unmoved, unproven service caps the whole tier.

1 · Procurement spec

"Internet-facing and crown-jewel services must be MARS-5; inner tiers stepped down by risk." A vendor cannot win the deal without proving its stars. The question enters the RFP and becomes a market force.

2 · Architecture baseline

Minimum rating per zone: internet-facing 5, DMZ 4, internal 3, crown jewels 5. Highest at the two high-risk ends, right-sized in between.

3 · Audit and regulatory artefact

"Show the external tier is MARS-5 with current evidence" — a measurable control, unlike "are you compliant?" A zone's rating is gated by its weakest external component, so it is auditable rather than a vanity badge.

4 · Board metric

One number a board reads instantly, like a credit rating or a crash rating: "what is our perimeter's MARS score?"

Worked example. An operator sets policy that its internet-facing tier and crown-jewel services must be MARS-5. It deploys a moving surface (L), operates its own servers so no patch is distributed plus runtime shielding and non-replay (R), a fleet control plane that re-keys on one detection (S), in-request response (O) and micro-segmentation (P). To earn and hold level 5 it runs an MRA-class adversary against its own perimeter continuously and records the measured results in tamper-evident form for independent verification. One unmoved, unproven service caps the tier.
Beyond enterprise IT

Built for cyber resilience and critical-infrastructure protection

Autonomous attackers do not stop at the enterprise. The systems that run essential services — energy, water, transport, health, finance — are increasingly internet-reachable and agent-exposed, and a breach there is measured in public safety, not downtime. MARS gives that world a measurable, provable resistance score.

Measurable resilience

MARS turns "are we resilient to AI attack?" into a graded, evidence-backed answer per zone — including the internet-facing OT and ICS gateways where moving-target defence and rapid containment matter most.

Evidence a regulator can check

Because high ratings are earned by tamper-evident, independently verifiable evidence, a MARS assessment produces the kind of artefact that cyber-resilience and critical-infrastructure regimes ask for: measured, dated and reproducible, not asserted.

Operator and board assurance

One number, per tier, that an operator and a board can act on — is the external surface of this essential service resistant to a machine adversary, and can we prove it today.

MARS is a measurement, not an accreditation. It complements sector cyber-resilience and critical-infrastructure frameworks rather than replacing them.

Honest by design

What MARS does not claim

A rating is only trustworthy if it states its own limits. MARS does, up front, so it cannot be misread.

Access tiers matter

Resistance that holds against an unauthenticated attacker may not hold against an authenticated agent. Every rating states its tier — a 4-star against outsiders is not a 4-star against a valid session.

Learn-resistance is not everything

It stops enumeration and replay. It does not, by itself, stop abuse of an endpoint the attacker is authorised to reach — that is a different layer.

Reuse-resistance splits by deployment

Strong for systems you operate (no patch is distributed, nothing to diff). Only partial for software you distribute or open-source, where the patch is public — there it raises the attacker's cost, it does not make weaponisation impossible. The real patch-to-exploit sample caught in 2026 was open-source, the hard case. MARS says so.

Where it sits

MARS does not replace what you already have

vs control catalogues

OWASP AI Exchange, AISVS, MITRE ATLAS, NIST AI RMF tell you what to build. MARS is the outcome rating that sits on top and grades whether, together, they actually resist the machine adversary. It consumes them; it does not compete.

vs CVSS

CVSS scores the severity of a specific flaw. MARS scores whether a system resists an autonomous attacker's attempt to find and weaponise flaws at all. They compose; neither substitutes for the other.

with ABCDE

The ABCDE model governs whether an acting agent can be trusted; MARS grades whether a system survives a hostile agent. Two halves of one coin, joined by the Evidence layer.

The companion model and the evidence stack

MARS does not stand alone

MARS grades whether a system survives a hostile agent. Its companion, the AgentPass ABCDE Model, governs whether an agent can be trusted in the first place. Between them they cover both sides of the agentic-security problem — the agent, and the system it attacks — and they are joined by a single shared layer: Evidence.

The AgentPass ABCDE Model

Five questions every AI agent must answer, continuously. Published by Raza Sharif, 2026 (DOI 10.5281/zenodo.23376107, CC BY 4.0).

A

Authenticate

Every agent carries a verifiable, attested identity. You know exactly which agent acted, and can prove it.

B

Behaviour

You know what normal looks like, so you catch what is not: drift, anomaly, and intent that does not match the task.

C

Control

Graded authority, enforced not advisory. A compromised agent is contained in seconds by revoking its identity.

D

Data

Provenance, lineage and integrity for what the agent takes in and the context it retains between interactions.

E

Evidence

A tamper-evident, independently verifiable record of what the agent actually did. The layer most frameworks assume and few build.

How the two interlock

ABCDE answers "can I trust the agent that is acting?" MARS answers "can the system survive a hostile agent?" The Evidence element is the hinge: it is how ABCDE proves an agent behaved, and how MARS proves a high rating holds. A, B, C and D are the price of entry; E is what turns "we think it behaved" into "we can prove what it did."

The evidence stack behind a rating

A MARS rating above level 2 must be proven, and the proof needs a trustworthy home. CyberSecAI builds that layer: agent identity and graded trust through AgentPass, an agent identity and trust platform, and a tamper-evident, independently verifiable record through the Agent Audit Trail. MARS consumes that evidence — it does not reinvent it.

The ABCDE model and MARS are both published openly under CC BY. AgentPass is a CyberSecAI platform; it is named here as one way to produce the evidence a high rating requires, not as a requirement of the rating. MARS grades the outcome and mandates no product.

Key terms

Glossary

Autonomous AI adversary

An AI that conducts offensive operations at machine speed, with no human in the decision loop, improving across attempts.

MARS Reference Adversary (MRA)

The versioned, tool-agnostic capability profile a test adversary must meet, so that ratings are comparable across targets and assessors.

Access tier

The position a rating is assessed from: Tier 0 unauthenticated, Tier 1 authenticated (a valid session is held), Tier 2 trusted supply-chain position.

Surface-recovery ratio

The fraction of real endpoints an enumerator recovers. Zero is the target for full Learn-resistance.

Moving-target defence

Changing the attack surface so it cannot be learned or reused. An established open field, not a single product.

Non-replayable

A captured request, token or exploit that cannot be used a second time — the core of Reuse-resistance.

Evidence ledger

A tamper-evident, independently verifiable record of a measurement, so a rating can be checked rather than trusted.

MARS-washing

Claiming resistance without measured proof. The evidence rule exists specifically to reject it.

Rating expression

A rating written with its conditions, for example MARS 4, MRA-1, Tier 1, as of 2026-11-18.

FAQ

Questions, answered plainly

Is MARS a certification I can get today?
Not yet. MARS is a published framework and an IETF Internet-Draft, with a working reference harness and the first measured ratings. A formal, accredited certification with an independent verifier is a later step. Today you can use it as a lens, a self-assessment, a procurement target and an architecture baseline.
Does a high MARS score mean I am unhackable?
No. MARS is a measurement against a defined adversary at a stated tier and point in time, not a guarantee against all future attackers. That is exactly why the top level requires continuous, not one-time, evidence, and why every rating carries its date and conditions.
Can any vendor reach the top, or is it rigged to one product?
It is deliberately a fair rating. Every axis is achievable with independently available technology — moving-target defence and zero-trust perimeters for L, non-replay and runtime shielding for R, fleet response for S, in-request enforcement for O, micro-segmentation for P. MARS grades the outcome and mandates no product.
My team ships open-source. Can we still score well?
On most axes, yes. The one honest caveat is R (reuse): because you must publish your patches, an AI can diff them, so you cannot win R through secrecy. You raise the attacker's cost with bundled or backported fixes, implementation diversity and runtime shielding — and you treat patch speed as a security metric, because the window from patch to exploit is now minutes.
How is a MARS score actually calculated?
Each of the five axes is scored 0, 1 or 2 from a measured result against the MARS Reference Adversary. The overall rating is the highest level for which all required axes score 2, gated by the weakest axis — never an average. Above level 2 the result must be reproducible and recorded in tamper-evident form.
What does the first measured run actually prove?
That the method runs and produces real, reproducible numbers: a static service rated MARS 0 (every route mapped, exploit reusable) and a moving-target service rated MARS 3 (nothing mapped, nothing replayable), with S and P honestly marked not assessed because they need a multi-node fleet harness. Every number was observed and written to a ledger that re-verifies.
Why does "Spread" use the letter P?
The five axes are Learn, Reuse, Scale, Outpace and Spread. Scale already takes the S, so Spread is keyed to its second letter, P. The axes are best remembered as a sentence rather than an acronym: an AI attacker cannot learn it, reuse against it, scale on it, outpace it, or spread through it.
Who maintains MARS and the reference adversary?
Both the reference-adversary profile and the scoring criteria are versioned and carry baseline dates, revised as attacker capability advances, with every rating citing the version it was earned under. The intent is for maintenance to pass to an independent body over time, as vehicle-safety and vulnerability-scoring standards have.
How do I cite or build on it?
The paper is on Zenodo (DOI 10.5281/zenodo.24341963) and the specification is IETF Internet-Draft draft-sharif-mars-rating, both under CC BY 4.0. You are free to use and adapt with attribution.