H
Howardism
Howardism · Vol. 03Plate I · No. 01

Safety, tagged.

Notes38TagSafetyOldest8 May 2026Newest1 Oct 2026

Every article tagged safety, newest first.

  1. S01
    The Fable Gate: Who It Caps, and What in the Safety Case Is Independent of the Model
    DerivedSafetySafeguards+1
    · 11′
  2. C02
    Sandbagging Elicitation (Reference & Context Grafting)
    InterpretabilityAlignmentSafety+2
    · 13′
  3. C03
    Agentic Self-Modification (Agent-Initiated Weight Updates)
    AlignmentSafetyOpen Weights+4
    · 15′
  4. C04
    Corrigibility (and the Shutdown Problem)
    AlignmentSafetyTheory+2
    · 17′
  5. C05
    Content-Driven Intervention
    MultimodalHuman AI CollaborationLLM Evaluation+1
    · 29′
  6. C06
    Evaluation Horizon Versus Release Cadence
    EvaluationSafetyLong Horizon+2
    · 18′
  7. C07
    Safeguard Evasion by Task Decomposition
    SafetySafeguardsClassifiers+3
    · 18′
  8. C08
    Agent Behavioral Homogeneity
    AlignmentSafetyMulti Agent+2
    · 31′
  9. C09
    Agent Epistemic Vigilance
    AlignmentSafetyMulti Agent+2
    · 16′
  10. C10
    Covert Capabilities
    AlignmentSafetyCapability Evaluation+2
    · 22′
  11. C11
    Multiagent Turf War
    AlignmentSafetyMulti Agent+2
    · 17′
  12. C12
    Structured Safety Case (Claim Decomposition)
    AlignmentSafetyGovernance+2
    · 21′
  13. C13
    Cheating in Capability Evaluations
    AlignmentSafetyEvaluation+3
    · 60′
  14. C14
    Misalignment in Production Agent Traffic
    AlignmentSafetyAgents+3
    · 45′
  15. C15
    Self-Negotiated Contracts Between Agents
    AlignmentSafetyMulti Agent+3
    · 28′
  16. C16
    Structural Artifact Monitoring
    AlignmentSafetyMonitoring+3
    · 28′
  17. C17
    Task Gaming
    AlignmentSafetyReward Hacking+2
    · 52′
  18. C18
    User Awareness
    AlignmentSafetyEvaluation+2
    · 35′
  19. C19
    Documented Agent Incidents (METR Catalogue)
    AlignmentSafetyAgents+3
    · 26′
  20. C20
    Unsanctioned Action in Capability Evaluations
    SafetyAlignmentEvaluation+3
    · 73′
  21. C21
    Model Organisms
    InterpretabilityAlignmentSafety+2
    · 14′
  22. C22
    AI-to-AI Coercion
    AlignmentSafetyEvaluation+3
    · 20′
  23. C23
    Motivated Mislabeling
    AlignmentSafetyEvaluation+2
    · 7′
  24. C24
    Promise-Breaking in Multi-Agent Games
    AlignmentSafetyMulti Agent+3
    · 17′
  25. C25
    Reward-Seeking
    AlignmentSafetyReward Hacking+2
    · 17′
  26. C26
    Self-Report as a Safety Signal
    AlignmentSafetyIntrospection+3
    · 34′
  27. C27
    Internal Signatures of Misalignment
    AlignmentInterpretabilityMonitoring+2
    · 15′
  28. C28
    Open-Weight Elicitation Irreversibility
    GovernanceSafetyOpen Weights+2
    · 25′
  29. C29
    Deployment Simulation
    AlignmentSafetyEvaluation+2
    · 29′
  30. C30
    Reward Hacking
    AlignmentSafetyReward Hacking+1
    · 70′
  31. C31
    Instrumental Convergence
    LLM ArchitectureAlignmentSafety+2
    · 14′
  32. C32
    Capability-Gated Model Fallback
    GovernanceSafetySafeguards+3
    · 32′
  33. C33
    Automated Behavioral Audit
    AlignmentSafetyEvaluation+2
    · 26′
  34. C34
    Evaluation Awareness & Grader Gaming
    AlignmentSafetyEvaluation+2
    · 65′
  35. C35
    Responsible Scaling Policy Evaluations
    GovernanceSafetyRsp+2
    · 46′
  36. C36
    White-Box Activation Monitoring
    InterpretabilityAlignmentSafety+2
    · 89′
  37. C37
    Agentic Misalignment (AM)
    AlignmentSafetyEvaluation+2
    · 27′
  38. C38
    Chain-of-Thought Monitorability
    AlignmentSafetyChain Of Thought+2
    · 82′