Glossary

Terms from A to Z

Aggregation resistance

The property of a control of taking effect even where an attack is decomposed into many individually legitimate-looking micro-steps. One of the four assessment criteria (Chapter 3.2).

AI – artificial intelligence

Rendered "KI" in German usage. This work uses "AI" throughout, to align with the MHC designations and the frameworks cited.

Attacker patience

Assessment criterion: does the protective effect rest on an attack being too laborious for an adversary? Structurally weakened under Mythos.

Capability decoupling

Assessment criterion: does the likelihood assumption depend on a historical correlation between actor and capability that Mythos has dissolved?

Crypto agility

The ability of a system to change cryptographic mechanisms at short notice, for example on compromise of an algorithm or during PQC migration.

FIDO2 / WebAuthn

FIDO Alliance authentication standards for phishing-resistant MFA. They anchor authentication cryptographically to the origin domain.

Friction-only

Category for controls whose classic or commonly encountered implementation form no longer provides sufficient standalone protective effect against Mythos attackers, because it rests on limited attacker capacity or on humanly manageable reaction times. Four controls in Chapter 6.

Frontier AI

Systems at the performance frontier of AI development. In public discourse, systems of the capability class that MRIS calls Mythos are called frontier AI or agentic AI.

MHC – Mythos Hardening Control

The category, introduced in this handbook, of a supplementary control that closes a Mythos-relevant effectiveness gap which ISO 27002:2022 covers only implicitly. Thirteen MHC in Chapter 9.

MRIS – Mythos-Resistant Information Security

"MRIS" (Mythos-Resistant Information Security) denotes the independent framework presented here; this handbook is its reference edition. Wherever "MRIS" appears below, the framework is meant, not this single document.

Mythos

In this handbook a class term: agentic frontier models with autonomous vulnerability discovery and exploit chaining, irrespective of vendor (including Claude Mythos/Fable 5, OpenAI Daybreak with GPT-5.5-Cyber, Microsoft MDASH). Originally the designation of the reference model Claude Mythos Preview (Chapter 2.5).

Not affected

Category for controls that are neutral with respect to Mythos-specific threats. 23 controls in Chapter 7.

Partially degraded

Category for controls whose protective effect is preserved but whose originally assumed strength falls under Mythos. 37 controls in Chapter 5.

PQC – post-quantum cryptography

Cryptographic methods resistant to attackers with quantum computing capability. NIST standardization (ML-KEM, ML-DSA, SLH-DSA).

Robust

Category for controls whose protective effect derives from a hard barrier that persists under Mythos. 29 controls in Chapter 4.

SBOM – software bill of materials

A machine-readable list of all software components of an artefact. Formats: SPDX, CycloneDX. Required under CRA Annex I Part II.

SLSA – supply-chain levels for software artifacts

A framework for build provenance with four maturity levels. Level 3+ recommended for critical build paths.

SOAR

Security orchestration, automation and response. The platform category for automated incident response workflows.

SPIFFE / SPIRE

The standard and reference implementation for workload identities. The basis for zero trust architectures in cloud-native environments.

Store-now-decrypt-later

A threat model in which encrypted data exfiltrated today is retained in order to decrypt it once quantum computing becomes available.

TEE – trusted execution environment

A hardware-supported, isolated execution area for confidential data processing. The basis for confidential computing.

Threat priority score

Threat priority score = threat relevance × business impact × exposure × control gap, each factor on a scale of 1 to 5. The threat priority score is a qualitative prioritization index, not a substitute for a formal risk analysis (ISO/IEC 27005) or an organization-specific risk matrix. Multiplying the factors makes relative action priorities visible; the value is ordinal, not metric — a score of 400 is not "twice as urgent" as 200 but serves ranking and triage.

Time compression

Assessment criterion: does the control presuppose human reaction time in a chain in which the attacker operates autonomously?

TLPT – threat-led penetration testing

Penetration testing based on real threat scenarios, anchored in DORA Art. 26/27 and the TIBER-EU framework.

UEBA – user and entity behavior analytics

A detection category based on behavioural baselines and ML-supported anomaly detection.

Zero trust

The architectural principle that no actor, device or network is trusted implicitly. Every transaction is verified explicitly. NIST SP 800-207.