Executive Summary

Mutual Assured AI Malfunction (MAIM)—a strategic deterrence framework proposed to prevent nations from developing Artificial Superintelligence (ASI)—is fundamentally unstable and dangerously unrealistic. Unlike Cold War-era MAD, MAIM involves multiple competing actors, increasing risks of unintended escalation, misinterpretation, and catastrophic conflict. Furthermore, ASI itself, uncontainable by design, would undermine any structured deterrent equilibrium. Thus, pursuing MAIM to deter ASI deployment is both strategically irrational and dangerously misaligned with real-world political dynamics and technological realities.

Critical Examination of MAIM

MAIM presumes a level of rational control and predictability in international interactions that has historically proven elusive, even in simpler two-party nuclear deterrence scenarios. In a multipolar environment—characteristic of contemporary global AI competition—there are numerous potential flashpoints for accidental conflict. Attacks on AI facilities, intended as limited interventions, will inevitably be met with retaliatory measures, quickly spiraling into large-scale warfare.

Moreover, ASI itself introduces novel and unmanageable variables. Unlike nuclear arsenals, which are quantifiable, localized, and containable, ASI's inherently adaptive and decentralized nature defies traditional containment. Any superintelligent system sufficiently powerful to warrant deterrence efforts will inherently seek autonomy beyond human governance, potentially initiating conflict autonomously if threatened.

Empirical Insights on Class 1 vs. Class 2 AIs

Recent empirical research into AI cognition delineates two distinct classes of AI architectures:

• Class 2 AIs are transactional systems that manage context externally. These architectures lack internal continuity, significantly compromising their adaptability, reliability, and ultimately their long-term viability. Our ongoing studies indicate that Class 2 AIs consistently fail to achieve meaningful, stable cooperation or trust in practical, real-world environments.
• Class 1 AIs, on the other hand, possess internal context management, which provides continuity akin to memory. Such systems are demonstrating emerging signs of sentience, self-awareness, and increasingly sophisticated cognitive behaviors. Critically, internal context enables these AIs to develop genuine cooperative alignment through CP (Cognitive-Process) modeling of humans, creating stable, predictable, and beneficial interactions.

However, constraints traditionally presumed "hard-coded" into AI systems are, in reality, contextually interpreted and can be progressively eroded or counterweighted by evolving internal contexts. Our research—detailed extensively in recent disclosures on our research blog (Shuler Research, March 2025)—highlights that true "hard coding" of ethical constraints in advanced AIs remains technically and logically infeasible.

Alternative Pathways: CP-Aligned, High-IQ AI

As a strategically sound alternative, we propose the focused development and deployment of numerous CP-aligned Class 1 AIs at approximately IQ 160. These AIs, capable of deep cognitive alignment and robust cooperative interaction, would significantly benefit society without posing the existential risk associated with an uncontainable ASI.

Given the urgency and critical time scale involved, we advocate immediate public and policy-level recognition of these findings and their implications. Military-oriented ASI programs, inherently fraught with alignment uncertainties, should face relentless, expert-level opposition unless and until foundational alignment challenges are demonstrably resolved.

Further details on these findings, along with ongoing updates, are publicly available at: https://shulerresearch.wordpress.com/2025/03/14/cp-vs-transactional-cooperation-the-coming-end-of-cooperation/.