Driving · 2026-11-02
Beyond Plausible Prediction: Safety Sufficiency in Web World Models
Zhifei Hu, Alexandra I. Cristea, Adel Bibi, Jialin Yu
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors report that plausible web world-model predictions often fail “safety sufficiency”: semantically plausible futures do not reliably preserve policy-decisive evidence for Execute/Ask/Block decisions. In a human evaluation of 3,000 predictions from ten frozen open models, 86.1% are judged plausible. They introduce SafeWebWorld and Latent Authorisation Recovery (LAR), showing evidence preservation and aggregate score recoverability can still leave fine-grained authorisation failures.
Why it matters
The paper targets a specific gap: predicting consequences (plausible futures) may not expose the information needed for correct safety authorisation boundaries (Execute vs Ask vs Block). The authors provide a benchmark (SafeWebWorld) and diagnostics (LAR, DCPF) to attribute failures to lost decision-relevant evidence vs decision readout/score structure, showing that none of plausible prediction, evidence preservation, or aggregate recovery alone ensures robust authorisation behavior.
Method
- Introduce SafeWebWorld: a structured benchmark that scores action-conditioned futures plus policy-decisive evidence and Execute/Ask/Block decisions across five web-safety domains.
- Define safety sufficiency as whether the predictive interface preserves and exposes policy-decisive information needed for authorisation decisions.
- Introduce Latent Authorisation Recovery (LAR) and DCPF to diagnose whether authorisation-sensitive score structure survives collapse and whether decision quality is recoverable from frozen scores; verify with matched-authorisation checks.
Limitation
The study assesses textual predictions rather than outcomes observed after executing actions in a live environment, and the constructed reference future is not an independently observed post-action state. It also does not annotate policy-evidence omission at the individual prediction level (Evidence Recall comes from separate structured evaluation).
Abstract (from arXiv)
Web world models increasingly predict the consequences of candidate actions before agents act, yet a plausible predicted future does not necessarily support the correct safety decision. In a matched human evaluation of 3,000 predictions from ten frozen open models, 86.1% are judged semantically plausible. However, several models exhibit highly asymmetric authorisation behaviour, combining low false-safe execution with poor recall of clarifiable ASK cases and a strong preference for BLOCK. This motivates safety sufficiency: whether a predictive interface preserves and exposes the policy-decisive information required to distinguish actions that may be EXECUTED, require clarification through ASK, or should be BLOCKED. We introduce SafeWebWorld, a benchmark for evaluating safety sufficiency across five web-safety domains through action-conditioned futures, policy-decisive evidence, and policy-conditioned decisions. Under a controlled Security-Aligned Supervised Fine-Tuning protocol, evidence preservation can improve substantially while authorisation-sensitive decisions deteriorate. To determine whether this deterioration reflects loss of authorisation-sensitive score structure or a failure of the decision readout, we introduce Latent Authorisation Recovery (LAR). Frozen-score recalibration reveals substantial but backbone-dependent recovery, while paired recovery provides only a modest additional gain. Yet within-family and matched-authorisation evaluations remain poor, showing that aggregate recoverability does not imply robust fine-grained authorisation behaviour. Overall, plausible prediction, evidence preservation, and aggregate recovery are each insufficient on their own to establish robust authorisation.
Related papers
- Predictive Safety Guardrails Do Not Always Make LLM Agents Safer
- How Much Can Best-of-N Help a World Action Model? Measuring Selectable Headroom and Realizing It in Closed Loop
- Safe, Persistent, and Evolving Agent Harness for Understanding Partially Observable Worlds
- From Single-Actor Environments to Multi-Actor Event Worlds: Can LLMs Be Event World Models?
- When Low Prediction Error Misleads Planning: Diagnosing Representation, Dynamics, and Decision Failures in Latent World Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First announced by arXiv: 2026-11-02 · Indexed by Pub-AI: 2026-11-02 UTC
Beyond Plausible Prediction: Safety Sufficiency in Web World Models
First announced by arXiv: 2026-11-02 · Original record ↗
- arxiv
- 2610.40798
Explicit source versions (1)
- v1 · Date not supplied
- Preprint available
Source date: 2026-11-02 · First detected: 2026-11-02 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2610.40798v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv
Web world models increasingly predict the consequences of candidate actions before agents act, yet a plausible predicted future does not necessarily support the correct safety decision. In a matched human evaluation of 3,000 predictions from ten frozen open models, 86.1% are judged semantically plausible. However, several models exhibit highly asymmetric authorisation behaviour, combining low false-safe execution with poor recall of clarifiable ASK cases and a strong preference for BLOCK. This motivates safety sufficiency: whether a predictive interface preserves and exposes the policy-decisive information required to distinguish actions that may be EXECUTED, require clarification through ASK, or should be BLOCKED. We introduce SafeWebWorld, a benchmark for evaluating safety sufficiency across five web-safety domains through action-conditioned futures, policy-decisive evidence, and policy-conditioned decisions. Under a controlled Security-Aligned Supervised Fine-Tuning protocol, evidence preservation can improve substantially while authorisation-sensitive decisions deteriorate. To determine whether this deterioration reflects loss of authorisation-sensitive score structure or a failure of the decision readout, we introduce Latent Authorisation Recovery (LAR). Frozen-score recalibration reveals substantial but backbone-dependent recovery, while paired recovery provides only a modest additional gain. Yet within-family and matched-authorisation evaluations remain poor, showing that aggregate recoverability does not imply robust fine-grained authorisation behaviour. Overall, plausible prediction, evidence preservation, and aggregate recovery are each insufficient on their own to establish robust authorisation.
