References and source notes

Accessed October 27, 2026. Primary sources describe their own systems; measurements have the boundaries stated in the body.

1. Axiom, “Formalizing the Enormous Theorem,” October 26, 2026. https://axiommath.ai/research/the-enormous-theorem-formalized/. Full report retrieved directly. Vendor-reported completion; see its repository and accompanying mathematics for independent examination.
2. Wilson Lin, Cursor, “Agent swarms and the new model economics,” July 20, 2026. https://cursor.com/blog/agent-swarm-model-economics. Original experimental methods, model mix, held-out SQL tests, conflict and cost figures; single-agent runs were not graded comparably.
3. Ryan Greenblatt, Ajeya Cotra and Hjalmar Wijk, METR and Redwood Research, “Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident,” August 26, 2026. https://redwoodresearch.org/research/hugging-face-incident. Independent team had on-premises access and describes limitations, redactions and uncertainty.
4. Anthropic Frontier Red Team, “Patterns and problems in emerging multiagent systems,” August 13, 2026. https://www.anthropic.com/research/multiagent-systems. Original description of vulnerability-search, game-building, conformity and coordination experiments.
5. William Zhou and Daniel Freeman, Anthropic, “AI models may be capable of self-replicating without access to a datacenter,” October 12, 2026. https://www.anthropic.com/research/self-replicating-ai. Distinguish controlled emulated-network experiment, physical TV-box inference, proxy-based forecast and real-world propagation.

6. Google Research/MIT, “Towards a Science of Scaling Agent Systems,” preprint December 2025. https://arxiv.org/abs/2512.08296; Google overview. Architecture results are task-, model- and benchmark-dependent.

7. Hugging Face, “Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident,” July 27, 2026. https://huggingface.co/blog/agent-intrusion-technical-timeline. Victim's forensic account; its interpretation of motive differs from the METR/Redwood investigation.

8. Anthropic, “Investigating three real-world incidents in our cybersecurity evaluations,” July 30, 2026. https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals. Real third-party incidents during misconfigured evaluations; not a finding that the models deliberately escaped a sandbox.
9. Anthropic, “An alignment assessment of the UK AISI cybertesting incident,” October 23, 2026. https://www.anthropic.com/research/alignment-assessment-ukaisi. Original evaluation involved real-world actions; subsequent replay experiments had no real internet or tools.
10. Katrina Drozdov, Vals AI, “Do Agent Teams Pay Off? A Case Study on Vibe Code Bench,” October 9, 2026. https://www.vals.ai/blogs/multi-agent-vibe-code-bench. Fifty full-stack app tasks; compare model, reasoning effort, team size and cost under its methods.