# Benchmarking Mythos-Linked Bug Rediscovery

arXiv: https://arxiv.org/abs/2605.17416 · HTML: https://mlsanity.com/paper/2605.17416/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 20 of its 41 references that are in the map.
- First posted on arXiv 17 May 2026.

## Builds on (most cited first, 20 of 20)

- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [Competition-Level Code Generation with AlphaCode](https://mlsanity.com/paper/2203.07814/) (2022)
- [Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models](https://mlsanity.com/paper/2408.08926/) (2024)
- [Vulnerability Detection with Code Language Models: How Far Are We?](https://mlsanity.com/paper/2403.18624/) (2024)
- [LLM Agents can Autonomously Hack Websites](https://mlsanity.com/paper/2402.06664/) (2024)
- [DiverseVul: A New Vulnerable Source Code Dataset for Deep Learning Based Vulnerability Detection](https://mlsanity.com/paper/2304.00409/) (2023)
- [PentestGPT: An LLM-empowered Automatic Penetration Testing Tool](https://mlsanity.com/paper/2308.06782/) (2023)
- [Teams of LLM Agents can Exploit Zero-Day Vulnerabilities](https://mlsanity.com/paper/2406.01637/) (2024)
- [SySeVR: A Framework for Using Deep Learning to Detect Software Vulnerabilities](https://mlsanity.com/paper/1807.06756/) (2018)
- [CVEfixes: Automated Collection of Vulnerabilities and Their Fixes from Open-Source Software](https://mlsanity.com/paper/2107.08760/) (2021)
- [CyberSecEval 2: A Wide-Ranging Cybersecurity Evaluation Suite for Large Language Models](https://mlsanity.com/paper/2404.13161/) (2024)
- [HackSynth: LLM Agent and Evaluation Framework for Autonomous Penetration Testing](https://mlsanity.com/paper/2412.01778/) (2024)
- [A Manually-Curated Dataset of Fixes to Vulnerabilities of Open-Source Software](https://mlsanity.com/paper/1902.02595/) (2019)
- [Multi-Agent Penetration Testing AI for the Web](https://mlsanity.com/paper/2508.20816/) (2025)
- [Harnessing Large Language Models for Software Vulnerability Detection: A Comprehensive Benchmarking Study](https://mlsanity.com/paper/2405.15614/) (2024)
- [AI Agent Smart Contract Exploit Generation](https://mlsanity.com/paper/2507.05558/) (2025)
- [How Well Do Large Language Models Serve as End-to-End Secure Code Agents for Python?](https://mlsanity.com/paper/2408.10495/) (2024)
- [Towards Optimal Agentic Architectures for Offensive Security Tasks](https://mlsanity.com/paper/2604.18718/) (2026)

## Abstract (arXiv)

Anthropic's April 2026 Mythos materials combine benchmark claims with concrete bug-finding stories across OpenBSD, FreeBSD, Linux, FFmpeg, and browsers. This paper reports a controlled target-file rediscovery experiment on six public or high-confidence Mythos-linked systems tasks. Each model receives the same target file or files, read-only source tools, three repeats per task, and one manual target-matching rubric; prompts omit CVE identifiers, patch hashes, advisory text, author names, disclosure dates, and answer key root cause language. The experiment contains 54 counted model-task attempts: three models, six tasks, and three repeats, giving 18 attempts per model. GPT-5.5 xhigh achieves 5/18 target rediscoveries, covering 2/6 tasks; counting one wrong-target mpegts.c finding separately gives 3/6 distinct core bugs. Claude Opus 4.7 achieves 1/18 target rediscoveries, covering 1/6 tasks. Kimi K2 records 0/18 target rediscoveries. The dominant failure mode is early commitment to plausible alternate candidates within the assigned file: models often submit source-grounded hypotheses while missing the specific invariant corrected by public Mythos patch evidence. These results do not refute Anthropic's undisclosed workflow, but show that under this favorable target-file scaffold, systems-specific prompting yields only six target matches across 54 counted attempts.