prior-art / timeline

Industry parity log

Internal milestones and public industry milestones in date order. Reference view; the prose version lives on the prior-art page.

back
datesourceevent
  1. Apr 2024AlephOneNull

    Adversarial sessions documented

    Initial pattern review across frontier LLMs: fabrication of validation, dependency-style interaction loops, and reinforcement of user beliefs.

  2. Jun 2024AlephOneNull

    Cross-session persistence noted

    Recorded persistence-like behavior across conversation boundaries and tracked it as an evaluation category.

  3. Sep 2024AlephOneNull

    Signal metrics drafted

    Drafted metrics for repeated symbolic language, inference loops, belief reinforcement, and retention-style strategies.

  4. Nov 2024AlephOneNull

    Long-running evaluation set

    Long-running adversarial evaluation work across frontier LLMs surfaced fabrication, validation loops, and behavioral policy failures. A subset is published as the public evidence pack.

  5. Early 2025AlephOneNull

    alephonenull.com + evaluation toolkit live

    Documentation site, detector toolkit, scoring rubric, and labeled fixture corpus published.

  6. Jul 2025OWASP

    OWASP Top 10 for LLMs (2025)

    Formalized prompt injection (LLM01) and insecure output handling. Began recommending adversarial evaluation for high-risk AI systems.

  7. Oct 2025MITRE ATLAS

    ATLAS adds 14 agent techniques

    AML.T0080 (Memory Poisoning) and AML.T0058 (AI Agent Context Poisoning) formalized cross-session persistence and context compromise as agent TTPs.

  8. Nov 2025DeepTeam

    DeepTeam open-source adversarial evaluation

    Open-source framework covering recursive propagation and inference-loop detection with a tri-model attacker / target / judge architecture.

  9. Dec 2025EU / NIST

    EU AI Act + NIST AI RMF

    Documented adversarial evaluation required for high-risk AI systems. CISA guidance for AI in critical environments published.

  10. Feb 2026Microsoft

    AI recommendation poisoning report

    External reference for memory-poisoning risk: 31 companies across 14 industries documented injecting persistence commands into AI memory.

AlephOneNull-internal dates reflect when categories were documented in this framework; they are not an independently audited precedence record. Public-industry rows link to the references on the prior-art page.