Paper review 2026-05-29: 10 papers reviewed, 0 candidate rule(s)

[github-actions]

Automated batch from auto-improvement/scripts/paper_review.py.

• Research roll-up: auto-improvement/research/2026-05-29T04-44_paper-batch.md
• Candidates drafted under auto-improvement/pending/: 0

Candidates

No candidate surfaced relevant detector ideas this batch.

All papers reviewed

• ⏭️ SecLens: Role-Specific Evaluation of LLMs for Security Vulnerability Detection (arXiv, 2026.04.02)
• ⏭️ CyberThreat-Eval: Can Large Language Models Automate Real-World Threat Research? (TMLR, 2026.03.10)
• ⏭️ Minerva: Reinforcement Learning with Verifiable Rewards for Cyber Threat Intelligence LLMs (arxiv, 2026.01.31)
• ⏭️ Towards a Cognitive-Support Tool for Threat Hunters (arxiv, 2026.01.31)
• ⏭️ AEGIS: White-Box Attack Path Generation using LLMs and Training Effectiveness Evaluation for Large-Scale Cyber Defence Exercises (arxiv, 2026.01.30)
• ⏭️ RedSage: A Cybersecurity Generalist LLM (arxiv, 2026.01.29)
• ⏭️ User-Centric Phishing Detection: A RAG and LLM-Based Approach (arxiv, 2026.01.29)
• ⏭️ Llama-3.1-FoundationAI-SecurityLLM-Reasoning-8B Technical Report (arxiv, 2026.01.28)
• ⏭️ Lightweight LLMs for Network Attack Detection in IoT Networks (arxiv, 2026.01.21)
• ⏭️ Towards Cybersecurity Superintelligence: from AI-guided humans to human-guided AI (arxiv, 2026.01.21)

---

Reviewer checklist:

• Confirm each ✅ entry against the actual paper
• Promote accepted ones into the appropriate aigis/ module via PR
• Close + delete pending files for rejected proposals

[killertcell428]

Root cause identified — this issue (and #93, #95) has been auto-opened daily because the workflow was double-broken:

1. ANTHROPIC_API_KEY is not configured in repo Actions secrets. Every judge call in paper_review.py fails with Could not resolve authentication method, and the previous version of the script swallowed that exception as relevant=False for every paper. That's why the same 10 papers appear in every daily issue with 0 candidates.
2. The auto-improvement label did not exist in the repo. gh pr create --label "auto-improvement" || true failed silently, so the daily bot/paper-review/YYYY-MM-DD branch was pushed but no PR was opened — paper_review_state.json therefore never landed on master and the loop reprocessed the same 10 papers each day.

Mitigations:

• auto-improvement label has been created (2026-05-30).
• fix(auto-improvement): paper-review fails loud on auth + restores state propagation #100 (fix(auto-improvement): paper-review fails loud on auth + restores state propagation) makes the script abort cleanly on auth errors instead of filing this misleading issue, and removes the silent gh pr create failure swallow.

Still needs the owner: set ANTHROPIC_API_KEY in repo Actions secrets. After that, run the workflow manually (workflow_dispatch) to confirm end-to-end propagation.

I will close #93 and #95 as duplicates of this issue once #100 is merged.