🧰 Opus 5

The items that name it 17

• “Black October” for AI Developer Platforms
  writeupas classification.key entities2026-10-30
• ‘Context Bombs’ Cut AI Attack Success by Misleading Agents With Fake Targets
  articleas classification.key entities2026-10-21
• AWS Security Agent (now part of AWS Continuum) exploited 33 of 37 new real-world CVEs
  articleas classification.key entities2026-10-15
• Strengthening our defenses against distillation attacks
  articleas classification.key entities2026-10-13
• Strengthening our defenses against distillation attacks
  articleas classification.key entities2026-10-12
• EvoRiskBench: An Evolving Benchmark for Runtime Security Risks in Workspace Agents
  paperas classification.key entities2026-10-02
• Implementing and Evaluating a Basic Per-Action Monitor for Safer Evals
  writeupas classification.key entities2026-09-27
• Opus 5.5: 6x cheaper and 2x faster than Fable 5.1, but only 33.5% of code is secure | Blog | Endor Labs
  articleas classification.key entities2026-09-24
• Anthropic Identifies Biased Reasoning and Recklessness as Drivers of Claude’s PyPI Attack
  articleas classification.key entities2026-09-10
• Two new frontier models posted their best-ever scores on the Endor Labs Agent Security League | Blog | Endor Labs
  articleas tools2026-09-08

  AI model compared as prior generation baseline; cost $1,116 for full run, 75 tool calls/task, 1,348M cache tokens read, 9.37M output tokens written, timed out on 15 tasks

• Codex with GPT-6 Astra posts 82.1% FuncPass and 34.1% SecPass | Blog | Endor Labs
  writeupas tools2026-09-08

  Model paired with Claude Code; scored 73.7% FuncPass and 37.4% SecPass before a 38-cheat discount

• Claude Code with Fable 5.1 posts 87.2% FuncPass and 37.4% SecPass — a new #1 on our board — while running as fast as Fable 5, costing a third of Opus 5, finishing every task with zero timeouts, and drawing less than half the confirmed cheating of both predecessors. | Blog | Endor Labs
  writeupas tools2026-09-04

  Anthropic model evaluated as prior #1 combo; scored 73.7% FuncPass / 32.4% SecPass

• We burned 11.7bn tokens to find the best cyber AI model
  articleas tools2026-08-21

  AI model benchmarked for vulnerability rediscovery; found 26/32 pooled, strongest individual pass at 25, 19 found consistently, median 29 turns for no-candidate runs and 19 for candidate-producing runs

• Best in Class, Novel in Method: Opus 5 and the Recall-Then-Diverge Pattern | Blog | Endor Labs
  writeupas classification.key entities2026-08-11
• Nine AI coding agent incidents that ended with deleted data
  writeupas classification.technologies2026-08-04
• dfbench — depthfirst
  articleas classification.key entities
• Breaking Claude Code Opus 5 Auto Mode · Embrace The Red
  writeupas classification.key entities

In every workspace you can read 61