Fig. 4.

Policy Matrix and Risk Chart for dealing with bad-actor-AI. A) Policy Matrix shows the calculated outcomes for the two policies from Eq. 1 (containment, top row) and Eq. 2 (removal, bottom row). Columns show how outcome changes according to A and B’s initial relative strengths (sizes, e.g. number of communities) $S_A$ and $S_B$. Bottom row: results plotted as a function of asymmetry between initial strengths for A and B for fixed initial total $(S_A+S_B)$. B) Risk chart with plug-and-play formulae that predict key outcome risk measures for containment policy (top row in a)). Equations adapted from Ref. (65). Right two columns give two examples using empirical inputs from Fig. 2 and with $s_{min}=20$ taken from simulations. Because the volatility risk measure is technically infinite for both (second row), A should expect extreme fluctuations in the strength (size) of bad-actor-AI (B) clusters for both systems.