Agents1Password SCAM Benchmark: Security Skills Cut Agent Failures 97%Help Net Security·high signalXBlueskyLinkedInCopy link1200-word security skill document reduces critical credential failures from 65 to 2. Claude Opus 4.6 scores 92%, cheapest agent security measure.SourceSource pageHelp Net Security↳ Follow the threadPolicy dependency / Stack layer'Do this as quickly as possible' repeatedly got a Claude session flagged by a corporate security directorr/ClaudeAIStack layer / Threat patternAFL++ fuzzed agent-written reimplementations of ten Linux utilities: fewer memory errors than the shipped versions, but more infinite loopsarXiv 2609.18298Stack layer / Threat patternEncoding the Expected Answer as Executable Code Keeps Data-Science Agent Evals Valid on Live DataarXiv 2609.16487Stack layer / Threat patternTypeSafe AI ships Jev, a model that returns typed probabilistic values instead of text, at $0.042 per million input tokens and free outputTypeSafe AIPolicy dependency / Stack layerCodex makes Code Mode wrappers transparent to Guardian policy and adds read-only policy to MCP tool requestsGitHubStack layer / Threat pattern37,623 provenance-labeled agent PRs: Codex code was reverted half as often as human code, Devin's 31% more, and Claude Code PRs waited 12.6 hours for first reviewarXiv 2609.17598Stack layer / Threat patternRaising a Stated Failure Probability From 10% to 70% Changes Whether Frontier Models Check the Evidence by at Most 21 PointsarXiv 2609.17865Policy dependency / Stack layerAnthropic Folds Cowork Into Claude and Ships Docs, Slides and Design as Editable SurfacesAnthropic (corroborated by TechCrunch, The Next Web and Dataconomy, all 2026-09-16/17)