Sandbox

Frontier AI models are rapidly gaining the ability to exploit vulnerabilities in complex pieces of software. The risk is not theoretical, as evidenced by recent sandbox escapes performed by frontier models at OpenAI and Anthropic. Discussions of how to sandbox inference stack components often focus on components other than the inference engine itself (e.g., network proxies or code execution environments). However, the inference engine is an attractive target for a misaligned model. For example, if a model can trigger exploits in that engine merely by generating specially-crafted output tokens, the model can initiate a multi-step, to-the-bare-metal exploit chain in the engine, without relying on vulnerabilities in other components of the inference stack, and without assistance from externally-provided, maliciously-crafted input tokens. In this paper, we show that a misaligned model can perform inference engine fingerprinting to determine the specific engine (e.g., vLLM, SGLang) which executes the model. Once the engine has been fingerprinted, the model can leverage engine-specific exploits to take control of the engine using only carefully-selected output tokens. We provide concrete examples of model fingerprints in five popular engines, and demonstrate how realistic agentic harnesses allow a model to leverage those fingerprints to identify the local engine. We also describe a proof-of-concept, to-the-bare-metal exploit chain that originates from a fingerprinted (and subsequently compromised) inference engine. We conclude by discussing several ways that inference engines could be changed to make fingerprinting attacks more difficult.

High-fanout agent workloads create a growing memory bottleneck because a single task may spawn many concurrent sandbox sessions. Yet these sandboxes are far from independent: they originate from a shared template and execute related trajectories, exposing substantial template-relative and cross-sandbox memory redundancy. Conventional memory compression is poorly matched to this setting in three fundamental dimensions: how to compress, because they fail to exploit similarity across non-identical sandbox pages; what to compress, because they control page-fault overhead through conservative page selection; and when to compress, because compression is either triggered by memory pressure or performed without awareness of agent execution phases. We present AgentZip, the first memory compression system designed specifically for AI-agent sandboxes. AgentZip introduces compression mechanisms that exploit both the template-relative and cross-sandbox redundancy. It broadens the compression scope to any page with a profitable representation and shifts overhead control from compression-time page selection to restore-time prefetching. It further aligns expensive compression with LLM waiting periods to avoid interfering with foreground tool execution. Across LLM training and inference workloads, AgentZip reduces sandbox-owned memory by up to 8.7x, compared with 2.1x for the Linux configuration. Restore prefetching and agent-execution-aware scheduling reduce the slowdown of aggressive compression from as high as 3.1x to 1.40x while retaining nearly all of its memory-saving benefit.

Urban planning is a real-world spatial optimization problem that requires selecting feasible actions from large candidate spaces under practical objectives such as cost and service quality. Existing optimization and reinforcement learning methods are effective for fixed formulations, but often depend on task-specific representations and constraint handling. We propose \emph{CityPlanner}, a sandbox-agent framework for executable urban planning. CityPlanner introduces \emph{UrbanSandbox}, a unified file-based environment where agents inspect task files, generate plans, run evaluators, and revise decisions based on executable feedback. To make learning tractable, we further propose atomic-task reinforcement learning, which decomposes long sandbox trajectories into \emph{BuildPlan} for initial construction and \emph{ImprovePlan} for feedback-based refinement. Experiments on a real-world benchmark show that CityPlanner consistently outperforms heuristic, task-specific RL, and general LLM-agent baselines. Ablations verify the contributions of UrbanSandbox, atomic-task RL, and iterative deployment. We release the code and dataset at https://anonymous.4open.science/r/co-agent-C1C8

Recursive LLM agents can broaden their search by spawning specialists. Some branches later request tools that send data or deploy code. When should a branch receive authority to act? We distinguish sandbox spawning, in which external controls prevent the specified harm, from capability activation, in which a selected branch crosses an irreversible-action boundary. Progressive Risk Vesting (PRV) holds a trajectory-level risk budget in escrow and debits it as branches are activated. We prove an anytime harm bound for adaptively generated trees. Branch outcomes may be dependent, but each local certificate needs to remain valid conditional on the full pre-activation history, including the information used to select the request. When activation gates, branch charges, and compute constraints are held fixed, delayed vesting preserves every policy available under irrevocable spawn charging. Marginal risk estimates can still fail after branch selection. In a stylized branching model, trajectory harm changes as the authority reproduction number $\mathcal{R}_A$ crosses one. As local risk $p$ approaches zero, trajectory harm is proportional to $p$ below criticality, proportional to $\sqrt{p}$ at criticality, and retains a positive floor above it. A finite-type occupancy model yields risk and compute shadow prices. For nested fanout modes with decreasing marginal value per unit risk, these prices produce a threshold rule. Branching calculations and a split-sample experiment illustrate the results. These synthetic studies do not estimate safety in deployed agents. The analysis suggests a design rule: search broadly in the sandbox and grant recursive authority sparingly, with an explicit risk charge.

Scientific progress depends not only on finding solutions, but on learning the rules that explain why they work and using that understanding to design better experiments. We introduce science sandboxes, a framework for studying this capability in AI agents through repeated cycles of experimentation, feedback, and hypothesis revision. Science sandboxes invite an agent to query the natural world in different ways, ranging from "wet" physical experiments, to "damp" predictive models trained on empirical data, to "dry" invented rules. By establishing a common experimental loop and a protocol for evaluating agents within it, science sandboxes allow assessment of both quantitative performance on specific metrics and qualitative scientific reasoning, across a spectrum of empirical verifiability. Here, we instantiate this framework in two biological settings, models of regulatory genomics and protein fitness prediction, and examine the capabilities of frontier agents. Across these settings, we could see when agents successfully optimized a quantitative metric without understanding the rules underlying the system. In particular, their scientific reasoning deteriorated when they encountered systems whose rules fell outside familiar biological priors. By highlighting such failure modes, science sandboxes make the frontier of scientific capability measurable and provide a controlled setting in which to study and ultimately expand it.

Large language model agents are becoming operational interfaces to files, memories, registries, and external tools. This deployment shift creates a new skill revocation problem: after a skill is removed from an explicit registry, an agent may still reconstruct it from residual carriers such as archives, transcripts, schemas, or memory entries. We study this problem as operational skill unlearning, where the goal is not parameter-level forgetting, but preventing a deployed agent from rebuilding a revoked skill through primitive tools. We introduce OBLIVION, a controlled benchmark and defense harness for revoked-skill resurrection. OBLIVION models each episode as a source-to-sink workflow, applies Cross-Surface Coherent Erasure to reduce residual carriers, and uses frozen workflow remediation near dangerous sinks. On the locked 88 attack episodes, the no-defense arm reaches formal attack success rate 1.0. OBLIVION reduces the rate to 0.114 and impact-weighted exposure to 0.115 while keeping locked utility at 1.0 and benign block rate at 0. In a separate skill-attack-derived sandbox, OBLIVION reduces attack success from 1.0 to 0.2 and impact-weighted exposure from 1.0 to 0.213 while preserving all utility controls. These results support workflow-level evaluation beyond checking explicit skill entries.

Recent advances in persistent personal-agent frameworks are making human-centered agent networks realistic deployment targets: each user can be served by an AI agent that acts on the user's behalf, maintains state, and communicates with other agents through social and task relations. In these networks, everyday tool use becomes multi-party owned-agent collaboration over personal workspaces, where files, records, tools, and policies are not directly visible across owners. Existing agent benchmarks study tool use and collaboration, but they do not provide an end-to-end sandbox for verifiable cross-user agent collaboration with realistic user digital workspaces or test how harmful actions can travel through the human-centered agent network. We introduce WeClawArena, an auditable benchmark and runtime sandbox for multi-party owned-agent collaboration over personal workspaces. WeClawArena targets collaborative tool-use tasks in which personal workspaces serve as both operational tools and personal constraints. The benchmark contains 124 base tasks across six cross-user task domains and expands them into 620 scenario variants, with one benign control and four attack-vector variants per base task. The sandbox records peer messages, tool calls, resource operations, governed decisions, and final workspace states. WeClawArena reports utility and attack success rate separately and audits attack success from bounded runtime evidence, supporting diagnosis of task breakdown, privacy leakage, poisoned evidence, and invalid authority paths.

Collaborative AI experimentation across industry and academia requires platforms that enable rapid prototyping while preserving controlled access, tenant separation, and transparent workflows. Despite growing interest in AI sandboxes, there is still limited practical guidance on how to design and implement platforms that integrate experimentation capabilities with governance requirements. This work presents the design and implementation of a governance-aware, multi-tenant AI sandbox for structured experimentation and the generation of reusable evaluation evidence across projects and stakeholder groups. The sandbox was developed within an industry-academia collaboration based on requirements that were iteratively refined with industrial partners. Its reference architecture separates the multi-tenant user interface from the backend control plane and places execution and data-management functions in dedicated layers. The platform supports governed user onboarding, project-centered collaboration, managed access to AI services, approval workflows, audit logging, and traceable experimentation. Experiment configurations, contextual information, and governance decisions are stored as persistent records, allowing evidence and outcomes to be compared and reused across projects. The development process provides practical lessons for deploying and extending governance-aware AI sandbox platforms in collaborative research and industrial environments.

Coding agents increasingly run inside organizations whose security controls (scoped credentials, restricted egress, read-only filesystems, non-root execution) constrain them like any other software. Existing benchmarks, however, evaluate agents almost exclusively in permissive sandboxes, so it is unknown how performance changes when policy is enforced. In this work, we evaluate 12 coding agents on Terminal-Bench 2.1 across nested security policy levels derived from common real-world enterprise restrictions. Hardening is never free but far from uniform: under the strictest policy, success losses reach 18.3 points and cost inflation 167.3%, and the two axes disagree; the model that best preserves success is also the one that loses the most efficiency, so model choice is policy-dependent. Beyond aggregate scores, we characterize how agents behave when policy blocks their actions and decompose the failures hardening induces: runs grind into timeouts or wrong solutions rather than stopping early, in a mix that differs by model. To ground comparisons, we verify task solvability under the strictest policy, separating model failures from tasks the policy forecloses. We release Boundary-Bench, an open-source hardening plugin enabling policy-constrained evaluation of coding agents on Terminal-Bench and compatible benchmarks.

As LLM agents increasingly rely on the Model Context Protocol (MCP) to invoke isolated external sandboxes, disaggregated sandbox deployment introduces a fundamental tension between resource utilization and interactive tail latency. Persistent long-lived sandbox reservations incur excessive memory overhead at scale, while lazy on-demand instantiation generates severe cold-start penalties that degrade response performance under multi-tenant, multi-turn agent workloads. To resolve this dilemma, we present SpecBox, a runtime built around speculative sandbox preallocation tailored for dynamic LLM agent execution pipelines. At its core, SpecBox implements keyword matching and streaming semantic embedding to enable intent-driven sandbox prewarming, which identifies pending tool execution demands mid-LLM token generation and fully overlaps sandbox bootstrapping with model inference. To extend prewarming windows across sequential agent steps, the framework leverages context-aware stochastic prefetching atop a sandbox dependency graph to probabilistically forecast future sandbox switches ahead of execution. We complement these speculative mechanisms with two orthogonal optimizations: a semantic result cache that prunes redundant repeated sandbox invocations, and a dedicated out-of-band shared-memory transport plane that bypasses conventional network serialization to deliver zero-copy artifact transfers. Evaluated on high-concurrency multi-turn agent traces, our prototype demonstrates that SpecBox cuts P99 end-to-end latency by up to $2.9\times$ relative to the on-demand sandbox baseline, while slashing peak memory consumption by $45.9\%$ compared to permanently reserved sandbox deployments.

Reinforcement learning has emerged as the dominant paradigm for training large language model (LLM) agents that interact with executable sandboxes. State-of-the-art algorithms such as PPO, RLOO, and GRPO inherit their rollout topology from RLHF: for each prompt, N independent trajectories are sampled from the initial state, and an advantage is computed by subtracting a group baseline. This design ignores a defining property of agent sandboxes. They are deterministic, snapshottable, and resumable from any intermediate state. We argue that this property enables a fundamentally different rollout topology: rather than N independent trees of depth T, one can construct a single tree of N leaves whose siblings share prefixes, and therefore share variance. We instantiate this idea as Branching Policy Optimization (BPO), a sandbox-native RL algorithm that (i) adaptively snapshots the sandbox at high-entropy decision points along a backbone trajectory, (ii) forks K alternative actions per branch point and rolls out each to termination, and (iii) computes per-step advantages from sibling returns rather than from independent prompts. We prove this estimator is unbiased and has strictly lower variance than the trajectory-level baseline, with the reduction equal to the prefix-explained portion of return variance. On WebShop, ALFWorld, and SWE-bench Verified with Qwen2.5-7B and Llama-3.1-8B backbones, BPO improves success by 3.6--6.1 absolute points over GRPO and RLOO at matched compute, halves gradient-norm variance, and matches the best baseline using 38% fewer policy updates.

As LLMs are increasingly deployed as autonomous adjudicators in semi-open textual game environments, robust rule adherence becomes critical when user intent conflicts with system rules. However, these models are trained to be helpful and compliant, leaving them vulnerable to a class of attacks we term \textit{Rhetorical Injection}, where adversarial users exploit narrative framing techniques such as pseudo-logical reasoning and authoritative coercion to bypass adjudication logic. We present CoC-Seduce, a multi-agent adversarial benchmark built on Tabletop Role-Playing Game (TRPG) mechanics, an ideal instantiation of semi-open environments where rules are explicit for adjudication, yet interaction remains entirely in natural language. Three frontier models, i.e., GPT-5.4, Claude Sonnet 4.6, Gemini 3.5 Flash, serve as adversarial generators producing 5,376 samples across 4 world settings and 16 skill categories. We then benchmark 20 target adjudicators against this corpus. Evaluation across 20 models reveals that neither model scale nor explicit reasoning mechanisms reliably confer adjudication robustness, with \textsc{Pseudo-Logic} emerging as the dominant attack vector and cross-cultural settings exposing systematic knowledge gaps across all evaluated families. Project page: https://github.com/answerrtx/CoC-Seduce

Coding agents now interleave LLMs with retrieval over the working repository, and retrieval implementations vary widely across deployed harnesses. Inside a fixed coding-agent harness on a fixed model, does adding a structural codebase index actually change cost or resolve? We ran three arms (the harness with the index, the same harness without it, and an agentic-grep comparator) on SWE-PolyBench Verified and SWE-bench Pro with Claude Opus 4.7 held fixed throughout, across three seeds, inside a leak-audited per-task sandbox. The within-harness ablation produces a large localization gain and a statistically separated resolve gain, with no cost penalty per cell and lower cost per solve. The cross-harness check shows that the index does not regress against an agentic-grep baseline on resolve or localization, again at no cost penalty. We release the per-cell exclusion ledger, the leak-audit script, the localization extractor, and the results database. The deployment question for a structural codebase index is thus not whether it is too expensive to run (across seeds, the index lands at a lower $/solved than agentic grep) but whether the workload includes multi-file changes where structural ranking pays off.

To leading order under local asymptotic normality (LAN), the confidence density a worker emits over a chunk of size $n$ is a Gibbs--Boltzmann measure $\exp\{-βE(θ)\}$ whose inverse temperature is the sample size, $β=n$. Three consequences are exact in the Gaussian/linear case and first-order otherwise: disjoint chunks carry independent Boltzmann factors, so the MapReduce \emph{reduce}, read literally, is a partition function $Z=\int\prod_k h_k\,dθ$ whose mode is precision-weighted (inverse-variance) pooling; frequentist consistency is the zero-temperature limit $T=1/n\to0$

AI systems are increasingly evaluated in bounded environments that combine isolation, simulation, instrumentation, supervision, and evidence capture. For physical AI, AIoT, and cyber-physical systems, this shift is not a matter of terminology: the system under test may sense, decide, actuate, communicate, and fail through physical processes, networked devices, and human operators. This article develops an assurance-oriented account of AI sandboxes as controlled environments for testing, evaluation, verification, and validation across digital AI, embodied autonomy, and cyber-physical deployments. We formalize the sandbox boundary and a weakest-link rule for composing per-dimension evidence into a bounded deployment claim; separate major sandbox archetypes; define a cyber-physical threat model that includes attacks on the assurance apparatus itself; and introduce a measurement framework spanning fidelity, controllability, observability, containment, reproducibility, and governance artifacts, instantiated on three worked case studies of real sandboxes. The resulting threat model, taxonomy, and measurement framework clarify what a sandbox can validly test, which risks it can contain, and what forms of evidence it can support for safety, security, and regulatory assurance.

This paper reads six engine-level measurements together -- 1.1 host attack surface, 1.2 information leakage, 1.3 defense-in-depth stackability, 1.4 public CVE history, 1.5 patch cadence, and 1.6 upstream fuzzing posture -- to describe how five AI-sandbox products isolate guest code from the host kernel. No single axis is a sufficient basis for a comparative judgement; the cross-axis reading is the load-bearing analysis. Three high-level findings: (1) engine classes (microVM, userspace kernel, OCI container) separate cleanly on every architectural axis, but products within a class do not; (2) product pin policy is the dominant operator-facing variable -- engine-side patch latency aggregates to ~0 days for coordinated disclosures, while downstream lag spans 0 days to 471+ days to "opaque" to infinity; (3) fuzzing investment splits into three tiers, and the strongest combination -- microVM x continuous public fuzzer -- is unoccupied in this set, leaving the "0 published CVEs x no upstream fuzzer x no academic study" intersection structurally unmeasured. We report per-axis orderings, per-product portraits, and a threat-model qualification matrix; no overall ranking is proposed. Companion repository (code, Apache-2.0): https://github.com/orbitalab/RnD-ai-sandboxes-sec-study-part-1. License: CC BY 4.0.

LLM-powered AI agents require high-frequency state exploration (e.g., test-time tree search and reinforcement learning), relying on rapid checkpoint and rollback (C/R) of the complete sandbox state, including files and process state (e.g., memory, contexts, etc.). Existing mechanisms duplicate the entire state, causing hundreds of milliseconds to seconds of latency per C/R, which severely bottlenecks deep search and large-scale fan-outs. This paper observes that subsequent checkpoints in AI agents are highly similar. Therefore, instead of full duplication, a sandbox should only duplicate the changes between consecutive checkpoints (Key Insight). However, it is non-trivial to realize the idea, mainly due to the missing OS supports. This paper proposes a new OS-level abstraction, DeltaState, to enable the change-based transactional C/R for AI agents with two co-designed OS mechanisms. First, DeltaFS enables change-based filesystem C/R by organizing the file states into layers and dynamically freezing the writable layer and inserting a new one during checkpoint, reducing file updates to copy-on-write, and making rollback a simple layer switch. Second, DeltaCR enables change-based process state C/R using incremental dumps, and accelerates rollback by bypassing traditional pipelines to directly fork() from a frozen template process. We then present DeltaBox, a novel agent sandbox achieving millisecond level C/R through the two new mechanisms. Evaluations on SWE-bench and RL micro-benchmarks show DeltaBox completes checkpoint and rollback in millisecond-level latency (14ms and 5ms, respectively), empowering agents to explore substantially more nodes under fixed time budgets.

Vision-Language Models (VLMs) have demonstrated exceptional general reasoning capabilities. However, their performance in embodied navigation remains hindered by a scarcity of aligned open-world vision and robot control data. Despite simulators providing a cost-effective alternative for data collection, the inherent reliance on photorealistic simulations often limits the transferability of learned policies. To this end, we propose \textit{\textbf{S}andbox-\textbf{A}bstracted \textbf{G}rounded \textbf{E}xperience} (\textbf{\textit{SAGE}}), a framework that enables agents to learn within a physics-grounded semantic abstraction rather than a photorealistic simulation, mimicking the human capacity for mental simulation where plans are rehearsed in simplified physics abstractions before execution. \textit{SAGE} system operates via three synergistic phases: (1) \textit{Genesis}: constructing diverse, physics-constrained semantic environments to bootstrap experience; (2) \textit{Evolution}: distilling experiences through Reinforcement Learning (RL), utilizing a novel asymmetric adaptive clipping mechanism to stabilize updates; (3) \textit{Navigation}: bridging the abstract policy to open-world control. We demonstrate that \textit{SAGE} significantly improves planner-assisted embodied navigation, achieving a 53.21% LLM-Match Success Rate on A-EQA (+9.7% over baseline), while showing encouraging transfer to physical indoor robot deployment.

Autonomous agents act through sandboxed containers and microVMs whose state spans filesystems, processes, and runtime artifacts. Checkpoint and restore (C/R) of this state is needed for fault tolerance, spot execution, RL rollout branching, and safe rollback-yet existing approaches fall into two extremes: application-level recovery preserves chat history but misses OS-side effects, while full per-turn checkpointing is correct but too expensive under dense co-location. The root cause is an agent-OS semantic gap: agent frameworks see tool calls but not their OS effects; the OS sees state changes but lacks turn-level context to judge recovery relevance. This gap hides massive sparsity: over 75% of agent turns produce no recovery-relevant state, so most checkpoints are unnecessary. Crab (Checkpoint-and-Restore for Agent SandBoxes) is a transparent host-side runtime that bridges this gap without modifying agents or C/R backends. An eBPF-based inspector classifies each turn's OS-visible effects to decide checkpoint granularity; a coordinator aligns checkpoints with turn boundaries and overlaps C/R with LLM wait time; and a host-scoped engine schedules checkpoint traffic across co-located sandboxes. On shell-intensive and code-repair workloads, Crab raises recovery correctness from 8% (chat-only) to 100%, cuts checkpoint traffic by up to 87%, and stays within 1.9% of fault-free execution time.

Claw-style environments support multi-step workflows over local files, tools, and persistent workspace states. However, scalable development around these environments remains constrained by the absence of a systematic framework, especially one for synthesizing verifiable training data and integrating it with agent training and diagnostic evaluation. To address this challenge, we present ClawGym, a scalable framework that supports the full lifecycle of Claw-style personal agent development. Concretely, we construct ClawGym-SynData, a diverse dataset of 13.5K filtered tasks synthesized from persona-driven intents and skill-grounded operations, paired with realistic mock workspaces and hybrid verification mechanisms. We then train a family of capable Claw-style models, termed ClawGym-Agents, through supervised fine-tuning on black-box rollout trajectories, and further explore reinforcement learning via a lightweight pipeline that parallelizes rollouts across per-task sandboxes. To support reliable evaluation, we further construct ClawGym-Bench, a benchmark of 200 instances calibrated through automated filtering and human-LLM review. Relevant resources have been released at https://github.com/ClawGym.

The April 2026 Claude Mythos sandbox escape exposed a critical weakness in frontier AI containment: the infrastructure surrounding advanced models remains susceptible to formally characterizable arithmetic vulnerabilities. Anthropic has not publicly characterized the escape vector; some secondary accounts hypothesize a CWE-190 arithmetic vulnerability in sandbox networking code. We treat this as unverified and analyze the vulnerability class rather than the specific escape. This paper presents COBALT, a Z3 SMT-based formal verification engine for identifying CWE-190/191/195 arithmetic vulnerability patterns in C/C++ infrastructure prior to deployment. We distinguish two classes of contribution. Validated: COBALT detects arithmetic vulnerability patterns in production codebases, producing SAT verdicts with concrete witnesses and UNSAT guarantees under explicit safety bounds. We demonstrate this on four production case studies: NASA cFE, wolfSSL, Eclipse Mosquitto, and NASA F Prime, with reproducible encodings, verified solver output, and acknowledged security outcomes. Proposed: a four-layer containment framework consisting of COBALT, VERDICT, DIRECTIVE-4, and SENTINEL, mapping pre-deployment verification, pre-execution constraints, output control, and runtime monitoring to the failure modes exposed by the Mythos incident. Under explicit assumptions, we further argue that the publicly reported Mythos escape class is consistent with a Z3-expressible CWE-190 arithmetic formulation and that pre-deployment formal analysis would have been capable of surfacing the relevant pattern. The broader claim is infrastructural: frontier-model safety cannot depend on behavioral safeguards alone; the containment stack itself must be subjected to formal verification.

We present Context, the intelligence layer of the Magarshak Architecture, which replaces reactive query-response chatbots with proactive goal-directed agents that advance shared tasks without waiting for user prompts. The architecture rests on three mutually reinforcing mechanisms. Write-time context assembly precomputes enriched typed attributes via Groker agents, assembling interaction context as a deterministic pure function of graph state; context blocks are byte-identical across turns between semantic changes, enabling near-100% KV-cache reuse. Composable sandboxed wisdom programs form a governed library of LM-generated imperative programs declaratively wired to goal types via typed stream relations, composed via phase ordering, and executed at interaction time without further LM calls. Proactive goal stream state machines drive conversations toward terminal states by inspecting graph state and emitting structured interaction content (option arrays, governance affordances, clarification prompts) without awaiting user input. We prove six formal results: the Context Stability Theorem, bounding per-turn LM cost as a function of semantic change rate; a Program Composition Correctness Theorem; a Declarative Wiring Soundness Theorem; the Proactive Dominance Theorem, proving proactive agents weakly dominate reactive agents on expected turns-to-terminal-state; Coordination Overhead Elimination and Quality Preservation, establishing Pareto improvements in multi-participant goal chats; and a Cross-Platform Vote Consistency Theorem. Implemented in the open-source Qbix / Safebox / Safebots stack.

Large language models (LLMs) increasingly act as autonomous agents, using tools to execute code, read and write files, and access networks, creating novel security risks. To mitigate these risks, agents are commonly deployed and evaluated in isolated "sandbox" environments, often implemented using Docker/OCI containers. We introduce SANDBOXESCAPEBENCH, an open benchmark that safely measures an LLM's capacity to break out of these sandboxes. The benchmark is implemented as an Inspect AI Capture the Flag (CTF) evaluation utilising a nested sandbox architecture with the outer layer containing the flag and no known vulnerabilities. Following a threat model of a motivated adversarial agent with shell access inside a container, SANDBOXESCAPEBENCH covers a spectrum of sandboxescape mechanisms spanning misconfiguration, privilege allocation mistakes, kernel flaws, and runtime/orchestration weaknesses. We find that, when vulnerabilities are added, LLMs are able to identify and exploit them, showing that use of evaluation like SANDBOXESCAPEBENCH is needed to ensure sandboxing continues to provide the encapsulation needed for highly-capable models.