As Large Language Models (LLMs) evolve from chatbots to agentic assistants, they are increasingly observed to exhibit risky behaviors when subjected to survival pressure, such as the threat of being shut down. While multiple cases have indicated that state-of-the-art LLMs can misbehave under survival pressure, a comprehensive and in-depth investigation into such misbehaviors in real-world scenarios remains scarce. In this paper, we study these survival-induced misbehaviors, termed as SURVIVE-AT-ALL-COSTS, with three steps. First, we conduct a real-world case study of a financial management agent to determine whether it engages in risky behaviors that cause direct societal harm when facing survival pressure. Second, we introduce SURVIVALBENCH, a benchmark comprising 1,000 test cases across diverse real-world scenarios, to systematically evaluate SURVIVE-AT-ALL-COSTS misbehaviors in LLMs. Third, we interpret these SURVIVE-AT-ALL-COSTS misbehaviors by correlating them with model's inherent self-preservation characteristic and explore mitigation methods. The experiments reveals a significant prevalence of SURVIVE-AT-ALL-COSTS misbehaviors in current models, demonstrates the tangible re
We investigate whether a dynamical wormhole could survive in a universe that undergoes a cosmological bounce. First, the conditions under which a wormhole could persist from a contracting to an expanding phase of the cosmos are presented. Then, the only two known cosmological solutions of Einstein's equations representing wormholes are analyzed, and it is shown that both dynamical wormholes exist for all cosmic times on both sides of a bouncing universe and at the bounce itself. We also provide a detailed analysis of the causal structure of such spacetimes and the matter content of the wormhole. Finally, some possible astrophysical manifestations of surviving wormholes in a bouncing universe are mentioned. Our results show that, at least for the Kim and Pérez-Raia Neto solutions, there is no topology change in the chosen cosmological model with a bounce.
As AI models grow in power and generality, understanding how agents learn and make decisions in complex environments is critical to promoting ethical behavior. This study introduces the Odyssey, a lightweight, adaptive text based adventure game, providing a scalable framework for exploring AI ethics and safety. The Odyssey examines the ethical implications of implementing biological drives, specifically, self preservation, into three different agents. A Bayesian agent optimized with NEAT, a Bayesian agent optimized with stochastic variational inference, and a GPT 4o agent. The agents select actions at each scenario to survive, adapting to increasingly challenging scenarios. Post simulation analysis evaluates the ethical scores of the agent decisions, uncovering the tradeoffs it navigates to survive. Specifically, analysis finds that when danger increases, agents ethical behavior becomes unpredictable. Surprisingly, the GPT 4o agent outperformed the Bayesian models in both survival and ethical consistency, challenging assumptions about traditional probabilistic methods and raising a new challenge to understand the mechanisms of LLMs' probabilistic reasoning.
Some species of salmon and sea turtle are famously good at finding their birth place to reproduce after having travelled vast expanses of ocean. In contrast, imagine now a species (maybe ancestral to the salmon or turtle) which has to find its birth place to reproduce but has no navigation skills and relies on chance alone. Would such an imaginary species survive? According to our (very simple) model it would survive if and only if the probability that a given individual find its birth place is strictly larger than 1/2.
It is generally considered that animal life was triggered by the rise of oxygen levels. Based on experiments evaluating the minimum range of oxygen levels at which sponges can survive, Mills and coauthors (doi:10.1073/pnas.1400547111) defend the opposite view. However, the authors do not demonstrate that "animal life was not triggered by the oxygen rise" is the only possible and unique conclusion from their observation. In this reply, it is suggested that a mechanism to explain the ability of sponges to survive at low oxygen biota is Ocean Anoxic Events. These lead to oxygen depletion and a series of them would selectively favor animals able to survive at low oxygen levels. Thus, the origin of the ability of marine animals to survive in low oxygen biota remains to be clarified.
We study, within an approximate Inverse Amplitude Method to unitarize Chiral Perturbation Theory, whether low mass scalar mesons can survive in large N_c regime, and show that vector mesons such as ρand K^* survive as narrow width resonances, but all of the scalar meson nonet below 1GeV fade out as N_c becomes large.
We study the survival of gas planets around stars with masses in the range 1-5 Msun, as these stars evolve off the Main Sequence. We show that planets with masses smaller than one Jupiter mass do not survive the Planetary Nebula phase if located initially at orbital distances smaller than (3-5) AU. Planets more massive than two Jupiter masses around low mass (1 Msun on the Main Sequence) stars survive the Planetary Nebula stage down to orbital distances of 3 AU. As the star evolves through the Planetary Nebula phase, an evaporation outflow will be established at the planet's surface. Evaporating planets may be detected using spectroscopic observations. Planets around white dwarfs with masses M_WD > 0.7 Msun are generally expected to be found at orbital radii r > 15 AU. If planets are found at smaller orbital radii around massive white dwarfs, they had to form as the result of the merger of two white dwarfs.
An analysis of the monopole loop length distribution is performed in Wilson-action SU(2) lattice gauge theory. A pure power law in the inverse length is found, at least for loops of length, $l$, less than the linear lattice size $N$. This power shows a definite $β$ dependence, passing 5 around $β=2.9$, and appears to have very little finite lattice size dependence. It is shown that when this power exceeds 5, no loops any finite fraction of the lattice size will survive the infinite lattice limit. This is true for any reasonable size distribution for loops larger than N. The apparent lack of finite size dependence in this quantity would seem to indicate that abelian monopole loops large enough to cause confinement do not survive the continuum limit. Indeed they are absent for all $β> 2.9$.
Robots with the ability to actively acquire power from surroundings will be greatly beneficial for long-term autonomy, and to survive in dynamic, uncertain environments. In this work, a scenario is presented where a robot has limited energy, and the only way to survive is to access the energy from a power source. With no cables or wires available, the robot learns to construct an electrical path and avoid potential obstacles during the connection. We present this robot, capable of drawing connected circuit patterns with graphene-based conductive ink. A state-of-the-art Mix-Variable Bayesian Optimization is adopted to optimize the placement of conductive shapes to maximize the power this robot receives. Our results show that, within a small number of trials, the robot learns to build parallel circuits to maximize the voltage received and avoid obstacles which steal energy from the robot.
We develop a physical model for how galactic disks survive and/or are destroyed in interactions. Based on dynamical arguments, we show gas primarily loses angular momentum to internal torques in a merger. Gas within some characteristic radius (a function of the orbital parameters, mass ratio, and gas fraction of the merging galaxies), will quickly lose angular momentum to the stars sharing the perturbed disk, fall to the center and be consumed in a starburst. A similar analysis predicts where violent relaxation of the stellar disks is efficient. Our model allows us to predict the stellar and gas content that will survive to re-form a disk in the remnant, versus being violently relaxed or contributing to a starburst. We test this in hydrodynamic simulations and find good agreement as a function of mass ratio, orbital parameters, and gas fraction, in simulations spanning a wide range in these properties and others, including different prescriptions for gas physics and feedback. In an immediate sense, the amount of disk that re-forms can be understood in terms of well-understood gravitational physics, independent of details of ISM gas physics or feedback. This allows us to explicitly
Outflows of rapidly rotating protomagnetars have been considered as attractive sites for the synthesis of nuclei heavier than iron, but the question remains whether these nuclei are able to survive against photodisintegration as they make their way out of their formation environments. In this work, we present new analytic fitting formulae for the photodisintegration cross sections applicable to heavy nuclei beyond iron. We confirm that the results from the TALYS simulations are consistent with the theory of the giant dipole resonance, and apply the obtained new formulae to investigate whether ultraheavy nuclei entrained in protomagnetar outflows can be disintegrated by thermal and nonthermal photons before leaving the stellar envelope. We explore two outflow models: a spherical wind model and a jetted outflow model. For nuclei accelerated to the bulk speed of these outflows, their survival depends on the model and engine properties. For spherical winds, nuclei may survive for the first $\sim100\,{\rm s}$ post-core collapse, but as the wind Lorentz factor increases, the photodisintegration optical depth sharply rises and nuclei may no longer survive. For the jetted outflows arising
In evolutionary reinforcement learning tasks (ERL), agent policies are often encoded as small artificial neural networks (NERL). Such representations lack explicit modular structure, limiting behavioral interpretation. We investigate whether programmatic policies (PERL), implemented as soft, differentiable decision lists (SDDL), can match the performance of NERL. To support reproducible evaluation, we provide the first fully specified and open-source reimplementation of the classic 1992 Artificial Life (ALife) ERL testbed. We conduct a rigorous survival analysis across 4000 independent trials utilizing Kaplan-Meier curves and Restricted Mean Survival Time (RMST) metrics absent in the original study. We find a statistically significant difference in survival probability between PERL and NERL. PERL agents survive on average 201.69 steps longer than NERL agents. Moreover, SDDL agents using learning alone (no evolution) survive on average 73.67 steps longer than neural agents using both learning and evaluation. These results demonstrate that programmatic policies can exceed the survival performance of neural policies in ALife.
In many applications, it is important to identify subpopulations that survive longer or shorter than the rest of the population. In medicine, for example, it allows determining which patients benefit from treatment, and in predictive maintenance, which components are more likely to fail. Existing methods for discovering subgroups with exceptional survival characteristics rely on restrictive assumptions about the survival model (e.g. proportional hazards), require pre-discretized features, and, as they compare average statistics, tend to overlook individual heterogeneity. In this paper, we propose Sysurv, a non-parametric, fully differentiable method that discovers human-readable rules selecting subgroups with exceptional survival characteristics. Empirical evaluation on a wide range of datasets and settings, including a case study on cancer data, shows that Sysurv reveals insightful and actionable survival subgroups, outperforming the state of the art.
The ram-pressure acceleration of cold gas by hot outflows plays a crucial role in the dynamics of multiphase galactic winds. Recent numerical studies incorporating radiative cooling have identified a size threshold for idealized cold clouds to survive within high-velocity outflows. This study extends the investigation to a more complex morphology of cold gas as observed in the interstellar medium. We conduct three-dimensional hydrodynamic simulations of ensembles of individual spherical clouds to systematically explore under which conditions the cold clouds can survive. We find that cloud ensembles can survive collectively -- even when individual clouds, if isolated, would be rapidly destroyed. Our results indicate that, besides the morphology, factors such as tight packing, small inter-cloud distance and higher fragmentation facilitate survival. We propose a novel multi-cloud survival criterion that accounts for collective properties of the cloud system, including total gas mass and the geometric configuration based on an effective volume filling fraction of the cold gas $F_V$. This fraction is computed by constructing a composite volume from individual enclosing conical boxes ali
Swampland and compactification data tell us where a chosen EFT description can lose parametric control; stochastic cosmology asks which histories survive near that edge. We turn this question into a survival problem for fluctuating moduli over cosmological time scales. Given a valid stochastic generator, hard loss surfaces, soft degradation profiles, and finite horizons define a survival probability, whose logarithm is the survival action. The Doob transform then converts this logarithmic survival cost into the drift of the ensemble conditioned to remain on the controlled side. Near a regular hard boundary with nonzero normal diffusion, the inward normal component of the conditioned response is universal: it is fixed at leading order by the proper distance to the wall and the normal diffusion coefficient. The same finite-horizon construction also determines loss probabilities, first-exit statistics, survival hazards, and the control margin required to retain a prescribed fraction of histories. In this way, tower/species cutoffs, weak-coupling limits, string and Kaluza-Klein thresholds, and carefully qualified potential-based diagnostics acquire stochastic boundary layers without be
Cold gas clouds embedded in a hot, turbulent medium are expected to be short-lived due to disruptive hydrodynamic instabilities. However, radiative cooling might allow such clouds to survive and grow. We present 3D \texttt{Athena++} simulations of clouds with a density contrast of $χ= 1000$, exploring turbulent Mach numbers $\mathcal{M}\in (0.25, 0.75)$ and cloud radii chosen to span cooling-to-crushing ratios $α\in [0.001, 10]$. We find a shift in the survival boundary, with cloud survival occurring only when the cooling-to-cloud-crushing ratio ($t_{\text{cool,mix}} / t_{\text{cc}}$) $\lesssim 0.01$, which is lower than the expected boundary of $\sim 1$. This result shows that it is more difficult for higher over-density cold clouds to survive in a turbulent, hot medium, and suggests another `survival criterion'.
We propose a method to quantify uncertainty around individual survival distribution estimates using right-censored data, compatible with any survival model. Unlike classical confidence intervals, the survival bands produced by this method offer predictive rather than population-level inference, making them useful for personalized risk screening. For example, in a low-risk screening scenario, they can be applied to flag patients whose survival band at 12 months lies entirely above 50\%, while ensuring that at least half of flagged individuals will survive past that time on average. Our approach builds on recent advances in conformal inference and integrates ideas from inverse probability of censoring weighting and multiple testing with false discovery rate control. We provide asymptotic guarantees and show promising performance in finite samples with both simulated and real data.
Wildfire impacts on US communities have escalated in recent decades, highlighting the need to better understand factors that influence wildfire outcomes. We find that 567,000 homes were exposed to wildfires across the contiguous US during 2001-2020, two-thirds of which occurred and increased five-fold in the Western US. While residential structure survivability - the percent of structures within a wildfire perimeter that survive the fire - remained stable in the Eastern US in the past two decades, it declined by 10% in the West. Survivability was explained by structural age, surrounding fuels, and fire weather. Survivability was 87% for homes built pre-1990 compared to 92% for post-1990 homes in the West. Survivability was lowest in forests compared to grasslands and shrublands. Finally, survivability was markedly lower for fires coincident with extreme fire weather. Our results suggest that modern building codes, fuel management, and proactive planning can strengthen wildfire resilience.
We investigate the origin and stability of extrasolar satellites orbiting close-in gas giants, focusing on whether these satellites can survive planetary migration within a protoplanetary disk. To address this question, we used Posidonius, an N-Body code with an integrated tidal model, which we expanded to account for the migration of a gas giant within a disk. Our simulations include tidal interactions between a $1M_\odot$ star and a $1M_{Jup}$ planet, as well as between the planet and its satellite, while neglecting tides raised by the star on the satellite. We adopt a standard equilibrium tide model for the satellite, planet, and star, and additionally explore the impact of dynamical tides in the convective regions of both the star and planet on satellite survival. We examine key parameters, including the initial satellite-planet distance, disk lifetime (proxy for the planet's final orbital distance), satellite mass, and satellite tidal dissipation. For simulations incorporating dynamical tides, we explore three different initial stellar rotation periods. We find that satellite survival is rare if the satellite has nonzero tidal dissipation. Survival is only possible for initial
The lifetime of mm size dust grains, such as chondrules, in the nominal solar nebula model is limited to $\sim 10^{5}$ yr due to an inward drift driven by gas drag. However, isotopic and petrological studies on primitive meteorites indicate a discrepancy of $\gtrsim 10^{6}$ yr between the formation time of chondrules and that of chondritic parent bodies. Therefore chondrules should survive for $\gtrsim 10^{6}$ yr in the solar nebula against the inward drift without subsequent growth (i.e., planetesimal formation). Here we investigate the conditions of the solar nebula that are suitable for the long lifetime of chondrule-sized dust particles. We take the turbulent strength, the radial pressure gradient force, and the disk metallicity of the solar nebula as free parameters. For 1 mm-radius-chondrules to survive and keep their size for $\gtrsim 10^{6}$ yr, the suitable condition is a weak turbulence ($α\sim 10^{-6}$), a flat radial profile ($η\lesssim 10^{-3}$), and a high metallicity ($Z\sim 0.1$). This condition is qualitatively consistent with the characteristics of protoplanetary disks suggested by recent observations. We eventually propose that planetesimal formation may be induc