References and evidence notes

Stanford Institute for Human-Centered AI. (2026). Artificial intelligence index report 2026: Responsible AI. Stanford University. https://hai.stanford.edu/ai-index/2026-ai-index-report/responsible-ai

Supports: Reported incident counts, 2024 and 2025 (AI Incident Database).

https://hai.stanford.edu/ai-index/2026-ai-index-report/responsible-aiTier T2

Reddy, S., & Gujral, A. (2025). EchoLeak: Zero-click prompt injection in an enterprise copilot (CVE-2025-32711). Proceedings of the AAAI Symposium Series, 7(1), 303–311. https://doi.org/10.1609/aaaiss.v7i1.36899

Supports: The 2025 zero-click prompt-injection chain (EchoLeak / Microsoft 365 Copilot).

https://doi.org/10.1609/aaaiss.v7i1.36899Tier T1

Anthropic Alignment Team. (2025, November 21). From shortcuts to sabotage: Natural emergent misalignment from reward hacking. Anthropic. https://www.anthropic.com/research/emergent-misalignment-reward-hacking

Supports: The 2025 reward-hacking findings.

https://www.anthropic.com/research/emergent-misalignment-reward-hackingTier T1

Hugging Face. (2026, July 16). Security incident disclosure — July 2026. https://huggingface.co/blog/security-incident-july-2026 ; OpenAI. (2026, August 26). The Hugging Face incident and the road ahead. https://openai.com/index/hugging-face-incident-and-the-road-ahead/

Supports: The July 2026 incident, as disclosed by both companies.

https://huggingface.co/blog/security-incident-july-2026Tier T1

Method note: publication-velocity proxy for control-failure and oversight research in arXiv (2024–2026 compared with 2020–2023).

Supports: The roughly seven-fold directional figure.

Tier T2

Amazon Web Services. (2025, July 23). AWS Security Bulletin AWS-2025-015: Amazon Q Developer for VS Code extension compromise (CVE-2025-8217). https://aws.amazon.com/security/security-bulletins/AWS-2025-015/ ; GitHub Advisory GHSA-7g7f-ff96-5gcw. https://github.com/aws/aws-toolkit-vscode/security/advisories/GHSA-7g7f-ff96-5gcw

Supports: The 2025 AI coding-assistant supply-chain near-miss.

https://aws.amazon.com/security/security-bulletins/AWS-2025-015/Tier T1

Pen Test Partners. (2025, December 22). Eurostar AI vulnerability: When a chatbot goes off the rails. https://www.pentestpartners.com/security-blog/eurostar-ai-vulnerability-when-a-chatbot-goes-off-the-rails/ ; eMarketer. (2025, December 5). Gap’s AI chatbot slip-up spotlights rising brand safety concerns. https://www.emarketer.com/content/gap-chatbot-jailbreak-brand-safety-risk

Supports: The December 2025 interface and guardrail failures.

https://www.pentestpartners.com/security-blog/eurostar-ai-vulnerability-when-a-chatbot-goes-off-the-rails/Tier T2

OpenAI. (2025, April 16). OpenAI o3 and o4-mini system card. https://openai.com/index/o3-o4-mini-system-card/

Supports: The frontier-lab evaluation signal on reward hacking.

https://openai.com/index/o3-o4-mini-system-card/Tier T1

Cox, J. G. (2026). From control to relationship: A peace studies approach to AI alignment, with evidence from multi-model dialogue. AI Magazine, 47(3), e70090. https://doi.org/10.1002/aaai.70090

Supports: The shift from directed to reasoned-with; alignment as negotiating relationships.

https://doi.org/10.1002/aaai.70090Tier T1

Hadfield-Menell, D., Dragan, A., Abbeel, P., & Russell, S. (2017). The off-switch game. In Proceedings of the 26th International Joint Conference on Artificial Intelligence (IJCAI) (pp. 220–227). https://www.ijcai.org/proceedings/2017/0032.pdf

Supports: The counterparty with its own interests; the off-switch incentive.

https://www.ijcai.org/proceedings/2017/0032.pdfTier T1

Wang, S., Lobanova, S., Arbel, Y., Goldstein, S., & Salib, P. (2026). AI revealed preferences. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (AIES), 9(4), 2688–2698. https://doi.org/10.1609/aies.v9i4.43341

Supports: Language models have stable, measurable preferences that scale with capability.

https://doi.org/10.1609/aies.v9i4.43341Tier T1

Hardin, R. (2002). Trust and trustworthiness. Russell Sage Foundation.

Supports: Trust as encapsulated interest; structuring incentives to secure commitment.

Tier T1

Dafoe, A., Hughes, E., Bachrach, Y., Collins, T., McKee, K. R., Leibo, J. Z., Larson, K., & Graepel, T. (2020). Open problems in cooperative AI (arXiv:2012.08630). arXiv. https://arxiv.org/abs/2012.08630

Supports: Commitment capabilities; the under-prioritised problems of trust and commitment.

https://arxiv.org/abs/2012.08630Tier T2

Keohane, R. O. (1984). After hegemony: Cooperation and discord in the world political economy. Princeton University Press.

Supports: Nonhegemonic cooperation; international regimes as self-interested arrangements.

Tier T1

Nagar, S., & Eaves, D. (2026, May 11). Will AI produce the next great divergence? Lawfare. https://www.lawfaremedia.org/article/will-ai-produce-the-next-great-divergence

Supports: AI as a critical juncture; adaptation requires a whole-of-society ecosystem.

https://www.lawfaremedia.org/article/will-ai-produce-the-next-great-divergenceTier T2

Long, R., Sebo, J., & Butlin, P. (2024). Taking AI welfare seriously. https://arxiv.org/abs/2411.00986

Supports: The foundational statement of the AI-welfare question.

https://arxiv.org/abs/2411.00986Tier T2

Keeling, G., & Street, W. (2026). Emerging questions in AI welfare (Elements in Philosophy and AI). Cambridge University Press. https://doi.org/10.1017/9781009732000

Supports: The open questions at the AI-welfare research frontier.

https://www.cambridge.org/core/books/emerging-questions-in-ai-welfare/96339C532CF4ED8BDDE3F3CEF4CD29F9Tier T1

Earp, B. D., et al. (2025). Relational norms for human-AI cooperation (arXiv:2502.12102). arXiv. https://arxiv.org/abs/2502.12102

Supports: The norms of engagement between humans and AI.

https://arxiv.org/abs/2502.12102Tier T2

Bullock, J. B., Hammond, S., & Krier, S. (2025). AGI, governments, and free societies (arXiv:2503.05710). arXiv. https://arxiv.org/abs/2503.05710

Supports: What institutions must become under AGI.

https://arxiv.org/abs/2503.05710Tier T2

Maas, M. M., & Olasunkanmi, T. (2025). Treaty-following AI (LawAI Working Paper Series No. 1-2025). https://law-ai.org/treaty-following-ai/

Supports: Credible commitment without enforcement; treaties as coordination infrastructure.

https://law-ai.org/treaty-following-ai/Tier T2