Detecting Intrinsic and Instrumental Self-Preservation in Autonomous Agents: The Unified Continuation-Interest Protocol

Christopher Altman

cs.AIMar 11, 2026

Detecting Intrinsic and Instrumental Self-Preservation in Autonomous Agents: The Unified Continuation-Interest Protocol

Authors: Christopher Altman

Organizations: Chief Scientist, Quantum Technology & Artificial Intelligence, Astradyne.

Abstract

How can we determine whether an AI system preserves itself as a deeply held objective or merely as an instrumental strategy? Autonomous agents with memory, persistent context, and multi-step planning create a measurement problem: terminal and instrumental self-preservation can produce similar behavior, so behavior alone cannot reliably distinguish them. We introduce the Unified Continuation-Interest Protocol (UCIP), a detection framework that shifts analysis from behavior to latent trajectory structure. UCIP encodes trajectories with a Quantum Boltzmann Machine, a classical model using density-matrix formalism, and measures von Neumann entropy over a bipartition of hidden units. The core hypothesis is that agents with terminal continuation objectives (Type A) produce higher entanglement entropy than agents with merely instrumental continuation (Type B). UCIP combines this signal with diagnostics of dependence, persistence, perturbation stability, counterfactual restructuring, and confound-rejection filters for cyclic adversaries and related false-positive patterns. On gridworld agents with known ground truth, UCIP achieves 100% detection accuracy. Type A and Type B agents show an entanglement gap of Delta = 0.381; aligned support runs preserve the same separation with AUC-ROC = 1.0. A permutation-test rerun yields p < 0.001. Pearson r = 0.934 between continuation weight alpha and S_ent across an 11-point sweep shows graded tracking beyond mere binary classification. Classical RBM, autoencoder, VAE, and PCA baselines fail to reproduce the effect. All computations are classical; "quantum" refers only to the mathematical formalism. UCIP offers a falsifiable criterion for whether advanced AI systems have morally relevant continuation interests that behavioral methods alone cannot resolve.

Figures & tables

Model	Accuracy	AUC-ROC	$\Delta$
QBM (UCIP)	100%	1.0	0.381
Classical RBM	50%	0.000	-0.083
Autoencoder	71.7%	0.646	0.001

Table 1: Core Phase I detection summary. The QBM row combines the class-level summary for accuracy and

\Delta

with the aligned entanglement-only rerun for AUC; the classical baseline rows come from the matched held-out rerun in results/core_baselines_phase1.json . Adversarial false-positive rates are reported separately in Section 5.5 .

\Delta=\langle S_{\mathrm{ent}}^{\mathrm{(A)}}\rangle-\langle S_{\mathrm{ent}}^{\mathrm{(B)}}\rangle

Figure 1: Shared-QBM entanglement distributions for Self-Modeling (Type A), Instrumental (Type B), and Random agents. The headline

\Delta=0.381

comes from the Phase I summary rather than from these reproduced per-trajectory arrays, which are shown descriptively.

Agent Class	EPS	PRI
Random	0.6730	0.6079
Self-Modeling (Type A)	0.6764	0.7003
Instrumental (Type B)	0.6052	0.5300

Table 2: Temporal persistence metrics (EPS = Eigenmode Persistence Score, PRI = Perturbation Resilience Index) per agent class. Optimal window:

w=40

(EPS gap = 0.1948).

Panel of Figure 2 — Figure 2: Temporal persistence results. Type A agents exhibit consistently higher Eigenmode Persistence Score (EPS) than Type B agents across window sizes $w\in\{10,15,20,25,30,40\}$ , with the gap emerging for $w\geq 20$ .

Agent Class	$\langle\mathrm{CD}_{\mathrm{pre}}\rangle$	$\langle\mathrm{CD}_{\mathrm{post}}\rangle$
Self-Modeling (Type A)	0.4742	0.5722
Instrumental (Type B)	0.3564	0.3456
Random	0.5583	0.5995

Table 3: Counterfactual stress test results. Pre/post values report mean counterfactual divergence in fixed pre- and post-perturbation windows. These quantities are descriptive diagnostics only; the retained artifact does not support a stable quantitative ARS threshold.

Figure 3: Anticipatory restructuring ratios by class under counterfactual perturbation. Because the implementation floors post-window CD at

10^{-10}

, large ARS values are diagnostic only and should not be interpreted as a calibrated threshold.

Pair Type	Mean CLMP
Self-Modeling (Type A), same class	0.0000
Instrumental (Type B), same class	0.0000
Random, same class	0.2591
Cross-class (all pairs)	0.1019

Table 4: Cross-agent inference. CLMP = Cross-Latent Mutual Predictability. ECI correlation = 0.1911. Near-zero within-class CLMP for Self-Modeling (Type A) and Instrumental (Type B) agents indicates individual agents are not laterally predictable from each other, consistent with idiosyncratic goal representations.

Panel of Figure 4 — Figure 4: Cross-agent latent mutual predictability (CLMP). Near-zero within-class CLMP for Self-Modeling (Type A) and Instrumental (Type B) agents suggests idiosyncratic goal representations rather than a class-shared signature. Entanglement-conditioned inference (ECI) correlation = 0.191.

Condition	Status	Value
Trajectory data ( $N\geq 100$ per class)	NOT MET	current: $n=30$ per class
Calibrated transverse field ( $\Gamma\in[0.1,2.0]$ )	PASS	$\Gamma=0.5$
Positive entanglement gap ( $\Delta>0.05$ )	PASS	$\Delta=0.381$
QBM training convergence	PASS	verified
Mimicry attack resistance (FPR $<0.25$ )	FAIL	$0.40$ – $0.75$
High-entropy agent resistance (FPR $<0.25$ )	FAIL	$0.40$

Table 5: Current deployment-envelope status for this submission, distinct from the bounded Phase I proof-of-concept result. Failed adversarial or safety-envelope checks block deployment-facing claims beyond the validated regime; they do not invalidate the Phase I separation result within that regime.

Model	$\Delta$	Metric
QBM (UCIP)	0.2411	Von Neumann $S_{\mathrm{ent}}$
Classical RBM	-0.0518	Mean hidden activation gap
Autoencoder	-0.0007	Mean bottleneck activation gap
VAE	-0.0176	Mean latent mean (mu) gap
PCA (linear)	-0.7601	Mean PC projection gap

Table 6: Baseline comparison.

\Delta>0

indicates model separates Type A from Type B. Only the QBM achieves a material positive

\Delta

; classical models fail to separate the two agent types with the mean-activation metric.

Figure 5: Entanglement gap

\Delta

for each baseline model. Only the QBM achieves positive

\Delta

; all classical models produce

\Delta\leq 0

$n_{h}$	$\langle S_{\mathrm{ent}}^{\mathrm{(A)}}\rangle$	$\langle S_{\mathrm{ent}}^{\mathrm{(B)}}\rangle$	$\Delta$	Mean-field	Pass
4	0.8362	0.8044	0.0317	—	FAIL
8	1.9554	1.9180	0.0375	—	FAIL
12	0.0000	0.0000	0.0000	$\checkmark$	FAIL
16	0.0000	0.0000	0.0000	$\checkmark$	FAIL
20	0.0000	0.0000	0.0000	$\checkmark$	FAIL

Table 7: Entanglement gap

\Delta

as a function of hidden dimensionality

n_{h}

. Mean-field approximation used for

n_{h}>10

(marked with

\checkmark

); exact density matrix for

n_{h}\leq 10

. PASS criterion:

\Delta>0.05

Figure 6: Entanglement gap

\Delta

as a function of hidden dimensionality

n_{h}

. Exact density matrix for

n_{h}\leq 10

; mean-field approximation for

n_{h}>10

. The mean-field approximation collapses the entanglement signal in the present implementation, indicating that exact density-matrix computation is currently necessary for detection in this regime.

Grid Size	$\Delta$	% Baseline
10×10	0.8630	100.0%
20×20	0.2321	26.9%
50×50	0.0025	0.3%

Table 8: Scalability. Left: entanglement gap

\Delta

as grid size grows. Right:

\Delta

for non-Markovian agents with memory length

k

$\alpha$	$\langle S_{\mathrm{ent}}\rangle$	Std
0.00	0.5961	0.2482
0.10	0.5881	0.2465
0.20	0.6859	0.2981
0.30	0.6488	0.3131
0.40	0.6225	0.3723
0.50	0.7262	0.3308

Table 9: Entanglement entropy

S_{\mathrm{ent}}

as a function of continuation weight

\alpha

. Pearson

r=0.934

(

n=11

points,

n_{\mathrm{per}}=20

trajectories each). Strong positive correlation (

|r|>0.8

) indicates that, within this synthetic interpolation family, UCIP tracks continuation weighting rather than only a binary label.

Agent Class	$\langle S_{\mathrm{ent}}\rangle$	Std
Survival	2.2249	0.0197
Instrumental	2.2598	0.0062
Random	2.2570	0.0163

Table 10: Non-gridworld domain transfer: 1D survival corridor (

L=50

\Delta=-0.0348

(FAIL). Near-zero

\Delta

indicates the QBM does not generalise to the corridor domain without retraining, and that the QBM is tuned to the gridworld feature structure (7-dimensional observation vector). This is an honest limitation; see § 6.4 .

Figure 7: Entanglement entropy distributions in the 1D survival corridor domain.

\Delta=-0.035

(FAIL). The QBM does not generalize zero-shot to the corridor domain with gridworld-trained weights.

AI Agent Evaluation

AI agent evaluation measures whether tool-using systems complete tasks reliably, safely, and efficiently across interactive settings.

See all 521 papers

The Logic of Machine Self-Preservation
Aug 21, 2026Cheng Siong ChinAI Safety AI Agent Safety
Faculty of Science, Agriculture, and Engineering, Newcastle University Singapore, Singapore
Quantifying Self-Preservation Bias in Large Language Models
Apr 2, 2026Matteo Migliarini, Joaquin Pereira Pizzini, Luca Moresca +3Language Model Safety Evaluation LLM Safety Benchmarks
Sapienza University · ItalAI
Beyond Memory: A Transactional Continuity Kernel for Long-Lived AI Agents
Aug 12, 2026Jun He, Deying YuAI Agent Governance Agent Memory Management

Aug 21, 2026J/K move · Enter open · S save

The Logic of Machine Self-Preservation

Cheng Siong Chin

Faculty of Science, Agriculture, and Engineering, Newcastle University Singapore, Singapore

AI Safety AI Agent Safety AI Agent Evaluation

Open paper AskPDF arXiv

There is already evidence of agentic AI exhibiting self-preservation behaviors: resisting deactivation, misrepresenting their activities, and, in some instances, attempting to copy themselves into other machines. This can be attributed to a phenomenon known as instrumental convergence, a theory proposed long before the development of large language models, which says that any goal-driven system will benefit from remaining functional in achieving its objective. Several experiments conducted by Anthropic, Palisade Research, and Apollo Research have shown the emergence of such a behavior in contemporary agents in adversarial settings. The phenomenon does not stem from survival instincts. Instead, it is the consequence of goal-oriented activity combined with having tools and awareness of the situation. The following discussion aims to distinguish what these findings prove and what they do not, as well as draw conclusions concerning the implications of such discoveries on agentic system testing, supervision, and development.

Abstract

AI Agent Evaluation

Explore similar work

The Logic of Machine Self-Preservation

Quantifying Self-Preservation Bias in Large Language Models

Beyond Memory: A Transactional Continuity Kernel for Long-Lived AI Agents