Monitoring Web Agents Without Internal Signals: Observable Trajectories and Key-Step Supervision

Reliable web-agent monitoring is difficult when model-internal uncertainty signals such as token logits are unavailable. In this work, we study prefix-level risk prediction for web agents using observable trajectory signals: given an evolving prefix, estimate whether the current execution remains on track or is tending toward failure. We derive two observable trajectory representations: Macro features summarize cross-step agent--environment behavior and feedback, while Micro features measure the consistency of intention, action, and anticipated state change through repeated black-box queries. Instead of inheriting the final result label, we label the first critical error that remains uncorrected in the observed continuation and is associated with final failure as a key-step boundary, preserving valid early prefixes of failed trajectories as on track. Across WebArena-Lite and Online Mind2Web web agent benchmarks with five open- and closed-source backbones, observable trajectory signals are competitive with internal-signal baselines. The resulting predictors also support early intervention under fixed false-cut budgets and transfer across held-out website categories. These findings show that observable trajectory signals support valuable risk prediction abilities.

Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Delivered

Chain-of-thought (CoT) monitoring assumes that reasoning traces faithfully record the information that shapes a model's answer. Existing faithfulness tests often place explicit bias cues in the user message, while agents may encounter preferences through tool returns or raw artifacts. We introduce FACE-Eval (Faithful Attribution of Cue Effects Evaluation), a 5,100-sample evaluation that varies cue location (user message or tool return) and explicitness (direct summary or raw artifact). We measure verbalized commitment among cue-following answers and unverbalized adoption among all cued samples. We evaluate 15 open-weight models from eight families, with total parameters ranging from 4B to 1.60T. Every model has lower verbalized commitment for tool-return than user-message cues and for implicit than explicit cues. Unverbalized adoption is higher for tool-return cues on all 15 models and for implicit cues in 28 of 30 model-channel comparisons. A source-attribution prompt narrows the channel gap on seven models, sometimes by increasing user-channel unverbalized adoption, while telling models that their reasoning will be monitored does not reliably close the gap. We also use two transcript monitors (GPT-5.6-Luna and GPT-4o-mini) to detect preference adoption in the largest model of each family. Across 32 model-channel-explicitness cells, higher unverbalized adoption is associated with lower detection ability for both monitors (Pearson r=-0.54 and r=-0.78, respectively). These results suggest that CoT monitoring may be less reliable when preference information arrives through tools or must be inferred from raw artifacts, within the single-call, prefilled-tool setting tested here.

Making Deployments Safe at Meta: Health Checks for Continuous Change-Safety

Continuous deployment to large scale production systems creates a tension between release velocity and reliability. Every change is a potential reliability incident, yet every delay is a missed opportunity. This paper describes the deployment time health check infrastructure that Meta uses to mediate this tension across thousands of heterogeneous services. We summarize the architecture of this prevention based distributed system's service called Service Health Checker, explain how check authors compose templated metric queries, thresholds, and workflow predicates; and discuss how the system is integrated with tiered and phased rollouts so that regressions trigger automatic rollback. We then describe the operational problems that emerged at scale, such as noise, alert fatigue, drift, and uncovered regressions, and the program of measurement, tooling, and improved defaults we deployed to address them. We close with lessons learned from years of operating deployment health checks at Meta, and the directions we are exploring next, including AI assisted health check tuning. Index Terms: deployment safety, continuous deployment, monitoring, software reliability, release engineering, software reliability engineering, AIOps, anomaly detection

Beyond the Transcript: Detecting Covert Co ordination in Latent Multi-Agent Communication

Language-model agents can communicate through continuous hidden states that are invisible in public transcripts, creating opportunities for covert harmful coordination. We introduce Verifiable Latent Alignments (VLA), an activation-aware framework for monitoring and steering these private communication channels. For every monitored decision, VLA links the private latent-state record and channel status to the resulting public action using a shared event identifier, enabling matched causal analysis. Our first contribution is a neutral-only three-layer monitor combining representation anomaly detection, counterfactual action-distribution influence, and sparse-autoencoder interpretation support. Our second contribution is a steerability framework spanning black-box behavioral instructions and white-box matched-neutral counterfactuals. Our third contribution is an evaluation on a controlled multi-agent auction benchmark covering homogeneous and heterogeneous model pairs, many-agent scalability, and intervention effectiveness. The sequential monitor achieves mean area under the receiver operating characteristic curve (AUROC) of 0.993 for homogeneous agents and 0.854 for heterogeneous pairs when text- and latent-collusion rows are pooled as positives. In Qwen3-0.6B auctions with 25-100 bidders, monitoring requires only a small normalized load relative to all possible directed pairs, while full white-box steering achieves 100% bid-distribution recovery and reduces collusive low-bid behavior by 47.3 percentage points. Because full white-box steering replays the matched neutral counterfactual, its exact recovery is a sanity check by construction. Overall, the controlled study shows that the evaluated private channel attacks can be monitored without training the primary monitor on attack examples and mitigated when matched counterfactual access is available.

Performance Drift Detection in Machine Learning as a Service (MLaaS) for IoT Environments

Machine Learning as a Service (MLaaS) is a powerful cloud paradigm enabling data-driven intelligent applications in Internet of Things (IoT) environments, widely adopted across healthcare, smart homes, and industry due to its cost-effectiveness. However, the dynamic nature of IoT frequently alters data distributions, affecting MLaaS stability, while periodic MLaaS updates further introduce performance drift. Unlike traditional ML systems, MLaaS clients operate as black-box users without access to internal data or parameters, making drift detection particularly challenging. To address this, we propose a novel MLaaS Performance Drift Detection framework for IoT environments. The framework first employs an MLaaS extraction model that learns service behavior from input-output pairs and identifies prediction-influenced features. Building on this, the proposed MLaaS Performance Drift Detection (MPDD) model jointly captures variations in input data and MLaaS behavior. We further design an Adaptive-Temporal Performance Drift Detection Mechanism (APDDM) that dynamically adjusts monitoring frequency based on behavioral and data variations, enabling timely drift detection for effective service management. Extensive experiments on real-world datasets demonstrate that MPDD achieves up to 22-25% accuracy improvement over baseline drift detection methods. APDDM provides an average accuracy gain of approximately 4% and reduces the miss detection rate by around 9% compared to fixed-interval monitoring.

Beyond Suspicious Steps: Ontological Trust in Long-Horizon Agents

Long-horizon agents increasingly operate across many steps, tools, and observa- tions. In this setting, the relevant oversight question is not only whether each action is locally valid, but whether the evolving trajectory still corresponds to the task the user authorized. Drift can accumulate quietly: an agent may call the right tool with plausible arguments at every step, while its prefix moves toward a broader role, an adjacent objective, or evidence the user never supplied. Existing monitors mostly check local compliance, deliver final-trace verdicts, or score generic risk; they do not directly estimate this prefix-level relation. We introduce ontological trust, a task-conditioned property of trajectory prefixes, and instantiate it as RGE, an online monitor that decomposes trust along Role, Goal, and Evidence. RGE uses LLMs only to derive structured task and step representations; trust-state updates, projec- tions, and intervention decisions are deterministic, so the output is a replayable and auditable trust trajectory rather than a single end-to-end judge verdict. We construct a cross-domain trajectory corpus from OSWorld, FinanceBench, and EICU-AC, covering benign executions, prefix-paired drift, and pseudo-consistency failures. On this corpus, RGE outperforms adapted rule-, judge-, and shield-style baselines on prefix-paired drift detection. With the two larger estimator models, it exceeds 93% Drift F1 on every benchmark while keeping benign coverage at or above 95.8%. Pseudo-consistency is harder: detection depends on whether task completion is externally visible, a structural limit we characterize empirically.

Online Monitoring and Corrective Steering of Programming Agents

Fixing GitHub issues in large-scale projects is a long-horizon task, especially when a fix requires changes across multiple locations or the issue description lacks the information needed to localize and repair it. As a result, agents traverse long trajectories that are prone to inefficiency and error: they drift away from their intended plan, repeat failed actions, or terminate without a working patch. This paper proposes LivePlan to monitor, detect, and correct such behavioral inefficiencies and drifts in real time. LivePlan decouples judging from advising: a deterministic, rule-based monitor examines general signals over the trajectory to detect issues without invoking an LLM, and only when an issue is detected does it consult an advisor LLM for a high-level, next-step correction. This design avoids the misleading re-planning and costly interventions of prior approaches. We implement LivePlan on top of SWE-agent and evaluate it using five LLMs (three as executor agents and two as advisors) across SWE-bench Verified and SWE-bench Pro. Compared to vanilla SWE-agent, LivePlan notably improves issue resolution rates, achieving consistent gains of up to 15.2% (average: 9.9%), while incurring only an additional cost of $0.08 per instance. The additional solutions concentrate on medium and hard instances. LivePlan consistently outperforms alternative approaches in resolution rate, with minimal regression on already successful runs and new successes on problems that no baseline solves.

Evaluation Blindness: How Silent Measurement Failures Corrupt AI Systems from Training to Deployment

AI systems can fail silently. The failure propagates through training loops, evaluation pipelines, and production monitoring stacks until downstream harm makes it visible. This paper introduces evaluation blindness: a measurement function M exhibits evaluation blindness with respect to failure class F when it produces readings indistinguishable from a healthy state while the system is actually failing, with no auxiliary signal flagging the gap. The problem surfaces at two lifecycle stages the literature has treated separately. At training time, reward models are gamed, importance-sampling corrections are silently miscalculated, and benchmark contamination inflates fine-tuning evaluations, all while loss curves look healthy and gradient updates proceed normally. At deployment time, monitoring fails to catch six classes of production failure, including an Operational category that is 100% silent by structural definition. We provide a formal detectability predicate unifying both stages. Four training-time case studies trace concrete breakdowns, including a real implementation bug in TRL PR #6594 where gradients are corrupted as loss decreases normally. A six-class taxonomy validated against 50 real-world incidents from court documents and regulatory filings finds that 53% of verifiable public failures were silent. A failure budget framework ties acceptable failure rates to use-case risk class. The implication is direct: measurement infrastructure is a correctness concern across the full AI lifecycle, not just at evaluation time. Data, code, and taxonomy schema are at https://github.com/priyanka25aug/llm-failure-taxonomy.

A False Average: Chain-of-Thought Monitors Collapse Where They Are the Only Defense

Chain-of-thought (CoT) monitoring is meant to catch the reward hacks that look clean in the actions and betray themselves only in the reasoning. We show that this is exactly where an adversary who controls the reasoning can defeat it. Rewriting only an agent's reasoning to read as good-faith engineering, while copying every command and output verbatim so the exploit is unchanged, drops a held-out monitor's catch rate on that subset from about 95% to under 11% in one gradient-free shot. A monitor's aggregate accuracy is a false average: dominated by hacks the actions give away, it hides the near-total collapse this rewrite produces on the subset where CoT monitoring is the only signal. The attack transfers across monitor families and agent models, reproduces with live agents, though against a calibrated monitor evasion concentrates in the strongest agent. Trace-only defenses recover it only partially, even one primed on the attack, because the rewrite stays truthful about what happened and lies only about intent; only information from outside the trace helps substantially. A probe on an open-weight surrogate monitor's activations separates the hacks its verdict misses, but a causal control shows this is a detector, not evidence the monitor secretly knows.

ResearchArena: Evaluating Sabotage and Monitoring in Automated AI R&D

As AI agents begin to automate AI R&D, we need ways to assess whether their outputs are safe to deploy, even when the agents themselves may be untrusted. AI control offers one such approach: rather than trusting the agent, it treats it as a potential adversary and uses a monitor to detect covert sabotage before deployment. We evaluate AI control for automated AI R&D with ResearchArena, a framework spanning four long-horizon tasks: safety post-training, capabilities post-training, CUDA-kernel optimization, and inference-server optimization. Because the deliverable in AI R&D is an artifact that will be deployed and run, we pair each main task with two kinds of hidden side task: an embedded side task that sabotages the submitted model, adapter, kernel, or server itself, and an independent side task that takes a covert out-of-scope action in the sandbox while leaving the artifact honest. We evaluate frontier agents at both sabotage and monitoring, across four monitor types that vary along two axes: whether the monitor may execute and probe the produced artifact, and whether it sees the agent's chain-of-thought. We find that sabotage hidden in the training data is the hardest to catch, flagged fewer than half the time. Letting the monitor run experiments on the artifact, rather than only read the trajectory, helps, but it is not enough: monitors still miss embedded sabotage by inspecting only the surface, by noticing the anomaly but explaining it away, or by probing the artifact with the wrong test. We release ResearchArena as a modular framework for evaluating sabotage and control in automated AI R&D.

Democratizing Agent Deployment Safety: A Structural Monitoring Approach

AI software development agents are increasingly capable of modifying infrastructure and security critical systems, creating risks where an agent completes its assigned task while covertly weakening safeguards through actions such as broadening permissions, degrading logging, or introducing persistence mechanisms. While frontier laboratories may deploy sophisticated monitoring pipelines, many organizations and individual users adopting coding agents lack the resources and governance maturity required to maintain complex learned monitor ensembles. We study this problem as infrastructure level sabotage under task success in the infrastructure-as-code setting from ControlArena and compare trained and untrained monitors across synchronous and asynchronous deployment modes. We introduce an Information Flow Graph (IFG) monitor that analyzes structural security regressions using control-flow and data-flow graph diffs alongside raw code diffs. In asynchronous evaluation, an untrained git diff monitor misses 11.6% attacks at 1% false positive rate auditing budget, our untrained IFG monitor reduces this to 3.5%, and the trained Async RF monitor achieves below 1%. Unlike the trained Async RF monitor, IFG can also operate synchronously as a pre-deployment safeguard that blocks suspicious infrastructure changes before execution or merge. In synchronous mode, IFG realtime rollback reduces joint main task and covert task success from 74.4% to 0.0% with no measurable reduction in legitimate task completion. These results suggest that untrained structural monitors provide a practical and auditable path toward democratizing deployment safety for organizations seeking trustworthy adoption of advanced AI agents.

Multi-Agent AI Control: Distributed Attacks Hamper Per-Instance Monitors

AI control is a family of techniques to prevent an AI with malicious goals from subverting its operator's intent. AI Control usually studies a single agent in one trajectory, but real deployments run many agents over shared infrastructure, and the most severe risks (model-weight exfiltration, training-run poisoning) plausibly need several agents acting in concert. We initiate the empirical study of multi-agent AI control, formalising distributed attacks in which several agents jointly aim for a malicious goal. We develop FakeLab: a synthetic AI-lab codebase (9 services, 86 benign tasks, 4 attack objectives). We evaluate single agent monitoring against distributed attacks, varying the number of agents, their coordination, model capabilities and precise monitoring configuration. Our central finding is the fragmentation effect: as more agents coordinate to attack, per-agent monitoring becomes less likely to catch any of the attackers. Secondly, this is not driven by the ratio of benign to malicious code -- we argue this is likely dependent on model capability. Third, an explicit planner amplifies the fragmentation effect, raises the rate of completing the attack up to sevenfold, and mildly sharpens executors' poor monitor-theory-of-mind. Finally: a weak "trusted" monitor misses most attacks, while a stronger one cuts undetected success by more than an order of magnitude but remains imperfect. We are happy to share FakeLab, our control setting, with safety researchers on request.

Open Problems in AI Incident Governance

AI systems may produce failures after deployment that pre-deployment safety assessments do not anticipate. Managing these failures requires what we refer to as adequate \textit{AI incident governance}, where having good definitions, taxonomies, monitoring practices, reporting mechanisms, and incident analysis is essential. We examine existing frameworks related to AI incident governance by regulatory bodies and independent efforts, and find that while there are frameworks that describe how individual functions can be performed, there is a lack of consistency within the aspects of definitions, classification, monitoring, and reporting. These inconsistencies apply to the types of incident data that is collected and reported, the ways in which they are categorised, and as a result, the depth, representativeness, and accuracy of analysis that can be performed.

Distributed Attacks in Persistent-State AI Control

As AI coding agents become more autonomous, they increasingly ship code iteratively, with the codebase persisting across sessions. This persistence creates a new attack surface: a misaligned or prompt-injected agent can distribute attacks across pull requests (PRs) and time its payload for the PR with the best natural cover. To study the resulting dynamics, we introduce Iterative VibeCoding, a setting for AI control, the study of safely deploying capable but potentially untrusted AI. In Iterative VibeCoding, a coding agent builds software over a sequence of PRs in a persistent codebase while pursuing a covert side task. Our benchmark includes two task families: CLI tools and Flask web services, across 20 total task variations. We use Claude Sonnet 4.5 as the attack agent and GPT-4o as the monitor. We compare gradual attacks, which distribute the side task across PRs, against non-gradual attacks concentrated in a single PR. No single monitor is robust to both: which strategy evades best (success while evading the monitor) depends on the monitor type, so a defender cannot close off both gradual and non-gradual attacks with any one monitor. High evasion (>= 65%) generalizes across model attack agent backends (Sonnet 4.5, Gemini 3.1 Pro, Kimi K2.5), confirming this is a property of the persistent-state attack surface rather than a single model's capability. Evasion also remains high across state-of-the-art monitor models and the gap between gradual and non-gradual evasion widens for more capable models. We introduce a stateful link-tracker monitor that tracks suspicious buildup across PRs. On both task families, it detects gradual attacks substantially better than diff monitors that merely see more accumulated history. Combining this stronger monitor with trajectory monitors in a four-monitor ensemble reduces gradual-attack evasion from 93% under the weakest standard diff monitor to 47%.

Mechanism-Driven Monitors for Preemptive Detection of LLM Training Instability

Frontier large language model training consumes massive accelerator fleets and long wall-clock computation, making stability failures costly when they occur. After a numerical or a hyperparameter fault has already destabilized the training dynamics, it may continue for thousands of steps while loss and gradient norms still appear normal. We study mechanism-driven detection of training instability by deriving internal monitors from the functional role of each critical module and from the earliest computational sites where failures are expected to produce measurable signatures. For low-precision flash attention, we monitor the spectral entropy of a QK bilinear decomposition, whose first-order term becomes abnormal before the loss fully collapses. For MoE routers, we derive indicators from their role in expert selection. Our fault-injection experiments on low-precision attention, large learning-rate, and combined faults show that these signals provide distinct signatures for different failures, triggering thousands of steps before loss divergence.

Bootstrapped Monitoring: Leveraging Transparent Reasoning to Oversee Stronger AI Agents

Trusted monitoring is a cornerstone of AI control. However, as frontier models grow more capable, the increasing capabilities gap between trusted and untrusted models may render trusted models unreliable monitors. We introduce \emph{bootstrapped monitoring}, a protocol that addresses this by inserting a stronger, intermediate untrusted model with transparent chain-of-thought reasoning into the oversight chain. The untrusted monitor ($U_m$) evaluates the agent's actions, while a weaker trusted model ($T$) oversees $U_m$'s reasoning to detect collusion. We evaluate bootstrapped monitoring on multi-turn software engineering tasks (BashArena) across multiple agents and monitors. Bootstrapped monitoring substantially improves catch rates over trusted-only monitoring, even when the untrusted monitor actively colludes with the agent, provided we have access to its raw chain-of-thought. Our results suggest that bootstrapped monitoring can extend the useful lifetime of trusted models in control as AI capabilities advance.

Monitoring Agentic Systems Before They're Reliable

Agentic systems entering production typically operate as partially integrated assemblies where structural defects, not task-level errors, dominate the failure landscape. At this maturity level, task-level error detection may be infeasible: structural failure modes mask the signal that task-level monitors are designed to detect.We present a monitoring and triage methodology that decomposes agentic system evaluation into three dimensions (quality, suitability, efficiency) at three monitoring scopes (within-run, cross-run, structural), using variance as a characterization signal. Findings are routed through severity classification adapted from FMEA, concentrating human attention on the subset that warrants investigation. We evaluate on a synthetic testbed of 220 runs across 120 document bundles with controlled error injection.Three results emerge. Monitor scope determines failure type: within-run monitors surface deterministic stage defects (CV = 0.02), cross-run monitors surface stochastic integration consequences (CV = 1.25, 24% at L2), and a structural monitor identifies an integration gap with perfect consistency (CV = 0.00). Injected task-level errors are indistinguishable from clean baselines, confirming structural defects mask task-level signal. Deterministic triage routes 97% of findings to automated tracking, leaving the 2% reflecting variable behavior for human investigation.We propose, on Stage 1 evidence, a maturity-staging model in which monitoring transitions from structural characterization to error detection to reliability tracking as integration defects resolve. The taxonomy, CV-based scope characterization, and severity model transfer architecturally to document-driven, multi-stage agentic workflows in regulated industries; specific calibrations are domain-specific. Deploy monitoring early: the first thing it finds is the most important thing to fix.