Safety, Security, and Cognitive Risks in World Models

2026-04-03T07:23:38Zabd7f3a137bbeaef104b8e36f9a316e27e03bdcf53d0de3460b243d957c17196
AI-assisted-verificationAssertainCWE-mappingEXHIB-benchmarkFoodGuard-4BFoodGuardBenchLLM-safetyNVDLA-case-studyRefinementEngineSelfGraderadversarial-attacksattacker-taxonomyautomated-assertionsbinary-function-similaritycyclic-Laplacedifferential-privacyfood-safetyguardrailshardware-securityjailbreak-detectionpolicy-deployment','network-enforcement','CTI-integration','façprivate-countsrepresentational-risktrajectory-persistenceworld-models

What happened

This collection aggregates 10 recent security- and safety-focused CS papers (arXiv, 2026-04-03) covering risks, benchmarks, and defenses across AI, hardware, privacy, and software supply chains. Key contributions include: (1) a survey of safety, security, and cognitive risks from learned world models with formal definitions (trajectory persistence, representational risk), an attacker-capability taxonomy, a unified threat model, and empirical trajectory-persistent adversarial attacks; (2) qualitative research on surveillance, censorship, and privacy harms from in-prison digital devices with UX-

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_cr
Record identifier
abd7f3a137bbeaef104b8e36f9a316e27e03bdcf53d0de3460b243d957c17196
Enrichment time
2026-04-03T07:23:38Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.