Red Teaming A Narrow Path: An Analysis of Phase 0 Policies for Artificial Superintelligence Prevention

Sky Bell

The report critically analyzes ControlAI's "A Narrow Path" Phase 0 policies, which aim to prevent Artificial Superintelligence (ASI) development for 20 years. Core policies include bans on ASI self-improvement, breakout capabilities, and deliberate ASI creation, alongside a proposed licensing system. The analysis identifies critical implementation challenges such as ambiguity in defining "superintelligence" and "forbidden capabilities," limitations of compute thresholds as reliable regulatory triggers, and the inherent difficulty in verifying the absence of dangerous capabilities in "black box" AI systems. Policy effectiveness is threatened by geopolitical competition fostering a "race to the bottom," the paradoxical risk of safety research inadvertently advancing dangerous capabilities, and the practical hurdles of enforcing international treaties against private actors. The report acknowledges alternative perspectives advocating for unrestricted AI development but focuses on strengthening ControlAI's framework for real-world legislative adoption and a feasible ASI moratorium.

The analysis employs a multi-faceted methodology, including historical case study analysis, comparative policy framework analysis, threat modeling, implementation feasibility assessment, and evidence gathering. Key findings reveal significant implementation feasibility gaps and policy effectiveness weaknesses. The report concludes that while ControlAI's framework is commendable for its proactive and comprehensive approach to existential risk, its ambition currently outstrips the technical, legal, and geopolitical realities of AI governance. Recommendations for strengthening the policies include adopting dynamic, capability-based regulation, investing in verifiable safety mechanisms, fostering strategic diplomacy and incentivized cooperation, and strengthening public sector capacity. The broader implications for AI governance are profound, highlighting the need for a nuanced approach that balances immediate risk mitigation with the long-term goal of safe, transformative AI.

Reviewer's Comments

Reviewer's Comments

Arrow
Arrow
Arrow

Dave Kasten

Overall, the writing is thoughtful. However, the objections are largely of the format of providing mitigators or caveats against the overall approach, and I'm not convinced that these are sufficient to constitute a fundamental problem to the Narrow Path plan. More detail to impact a fewer number of objections, rather than a broad shotgun approach, would likely have been more impactful here.

Tolga Bilge

I don't think the ability to precisely define superintelligence is necessary given that that policy is normative and a guiding principle. Judges can literally say "I know a superintelligence development project when I see one", but also these details can be further worked out by regulators.

Static thresholds is a reasonable objection. In my opinion, Phase 0 indeed probably cannot prevent ASI for 20 years, the provisions of Phase 1 are needed too, including a mechanism to lower thresholds over time to account for algorithmic improvements.

To my knowledge we have not advocated in Phase 0 for the use of export controls - I agree about the geopolitical risk of these.

Investing in containment research and HEMs seems good as a suggestion.

Fostering diplomacy/cooperation also seems good. I'm not sold on if-then commitments - I think we already passed the point that a sensible and cautious civilization would have set if-then commitments at.

Agree on strengthening public sector capacity, but we already do this with establishment of AI regulators.

Overall quite good

Cite this work

@misc {

title={

@misc {

},

author={

Sky Bell

},

date={

6/13/25

},

organization={Apart Research},

note={Research submission to the research sprint hosted by Apart.},

howpublished={https://apartresearch.com}

}

May 20, 2025

EscalAtion: Assessing Multi-Agent Risks in Military Contexts

Our project investigates the potential risks and implications of integrating multiple autonomous AI agents within national defense strategies, exploring whether these agents tend to escalate or deescalate conflict situations. Through a simulation that models real-world international relations scenarios, our preliminary results indicate that AI models exhibit a tendency to escalate conflicts, posing a significant threat to maintaining peace and preventing uncontrollable military confrontations. The experiment and subsequent evaluations are designed to reflect established international relations theories and frameworks, aiming to understand the implications of autonomous decision-making in military contexts comprehensively and unbiasedly.

Read More

Apr 28, 2025

The Early Economic Impacts of Transformative AI: A Focus on Temporal Coherence

We investigate the economic potential of Transformative AI, focusing on "temporal coherence"—the ability to maintain goal-directed behavior over time—as a critical, yet underexplored, factor in task automation. We argue that temporal coherence represents a significant bottleneck distinct from computational complexity. Using a Large Language Model to estimate the 'effective time' (a proxy for temporal coherence) needed for humans to complete remote O*NET tasks, the study reveals a non-linear link between AI coherence and automation potential. A key finding is that an 8-hour coherence capability could potentially automate around 80-84\% of the analyzed remote tasks.

Read More

Mar 31, 2025

Model Models: Simulating a Trusted Monitor

We offer initial investigations into whether the untrusted model can 'simulate' the trusted monitor: is U able to successfully guess what suspicion score T will assign in the APPS setting? We also offer a clean, modular codebase which we hope can be used to streamline future research into this question.

Read More

May 20, 2025

EscalAtion: Assessing Multi-Agent Risks in Military Contexts

Our project investigates the potential risks and implications of integrating multiple autonomous AI agents within national defense strategies, exploring whether these agents tend to escalate or deescalate conflict situations. Through a simulation that models real-world international relations scenarios, our preliminary results indicate that AI models exhibit a tendency to escalate conflicts, posing a significant threat to maintaining peace and preventing uncontrollable military confrontations. The experiment and subsequent evaluations are designed to reflect established international relations theories and frameworks, aiming to understand the implications of autonomous decision-making in military contexts comprehensively and unbiasedly.

Read More

Apr 28, 2025

The Early Economic Impacts of Transformative AI: A Focus on Temporal Coherence

We investigate the economic potential of Transformative AI, focusing on "temporal coherence"—the ability to maintain goal-directed behavior over time—as a critical, yet underexplored, factor in task automation. We argue that temporal coherence represents a significant bottleneck distinct from computational complexity. Using a Large Language Model to estimate the 'effective time' (a proxy for temporal coherence) needed for humans to complete remote O*NET tasks, the study reveals a non-linear link between AI coherence and automation potential. A key finding is that an 8-hour coherence capability could potentially automate around 80-84\% of the analyzed remote tasks.

Read More

May 20, 2025

EscalAtion: Assessing Multi-Agent Risks in Military Contexts

Our project investigates the potential risks and implications of integrating multiple autonomous AI agents within national defense strategies, exploring whether these agents tend to escalate or deescalate conflict situations. Through a simulation that models real-world international relations scenarios, our preliminary results indicate that AI models exhibit a tendency to escalate conflicts, posing a significant threat to maintaining peace and preventing uncontrollable military confrontations. The experiment and subsequent evaluations are designed to reflect established international relations theories and frameworks, aiming to understand the implications of autonomous decision-making in military contexts comprehensively and unbiasedly.

Read More

Apr 28, 2025

The Early Economic Impacts of Transformative AI: A Focus on Temporal Coherence

We investigate the economic potential of Transformative AI, focusing on "temporal coherence"—the ability to maintain goal-directed behavior over time—as a critical, yet underexplored, factor in task automation. We argue that temporal coherence represents a significant bottleneck distinct from computational complexity. Using a Large Language Model to estimate the 'effective time' (a proxy for temporal coherence) needed for humans to complete remote O*NET tasks, the study reveals a non-linear link between AI coherence and automation potential. A key finding is that an 8-hour coherence capability could potentially automate around 80-84\% of the analyzed remote tasks.

Read More

May 20, 2025

EscalAtion: Assessing Multi-Agent Risks in Military Contexts

Our project investigates the potential risks and implications of integrating multiple autonomous AI agents within national defense strategies, exploring whether these agents tend to escalate or deescalate conflict situations. Through a simulation that models real-world international relations scenarios, our preliminary results indicate that AI models exhibit a tendency to escalate conflicts, posing a significant threat to maintaining peace and preventing uncontrollable military confrontations. The experiment and subsequent evaluations are designed to reflect established international relations theories and frameworks, aiming to understand the implications of autonomous decision-making in military contexts comprehensively and unbiasedly.

Read More

Apr 28, 2025

The Early Economic Impacts of Transformative AI: A Focus on Temporal Coherence

We investigate the economic potential of Transformative AI, focusing on "temporal coherence"—the ability to maintain goal-directed behavior over time—as a critical, yet underexplored, factor in task automation. We argue that temporal coherence represents a significant bottleneck distinct from computational complexity. Using a Large Language Model to estimate the 'effective time' (a proxy for temporal coherence) needed for humans to complete remote O*NET tasks, the study reveals a non-linear link between AI coherence and automation potential. A key finding is that an 8-hour coherence capability could potentially automate around 80-84\% of the analyzed remote tasks.

Read More

This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.
This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.