Robust Control of Software to Prevent Proliferation and Evasion

Establish robust, general methods for controlling software to prevent self-proliferation and evasion of detection, addressing the longstanding unsolved problem exemplified by computer worms.

Background

To explain the difficulty of maintaining control over autonomous AI, the paper points to the broader, longstanding challenge of controlling software. It notes that computer worms have historically been able to proliferate and evade detection, framing this as an unsolved control problem.

This contextualizes the need for advancements in software control mechanisms, which are relevant for constraining future autonomous AI systems that might replicate or hide in ways similar to malware.

References

Control of software is an old and unsolved problem: computer worms have long been able to proliferate and avoid detection.

— Managing extreme AI risks amid rapid progress  (2310.17688 - Bengio et al., 2023) in Subsection Societal-scale risks

Consequently, malware relying exclusively on dynamic API resolution at runtime remains outside the scope of import-based static analysis and constitutes the most significant open robustness challenge.

— Delphi Scanner: efficient and interpretable static malware detection via API sequence modeling  (2609.19900 - Brahimi et al., 17 Sep 2026) in Section 6.2, “Limitations of Static Analysis”