{"id":"51e7cb39-794d-4f20-b917-5e01b3cb0389","arxiv_id":"2607.00186","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Dual-smartphone side-channel attack reconstructs 3D printer G-code from acoustic and magnetic emissions at 98.89% command-level accuracy in a 60 cm non-line-of-sight setup.","lead":"Researchers used two smartphones 60 cm away in a non-line-of-sight setup to capture acoustic and magnetic emissions from a 3D printer and reconstruct its G-code commands at 98.89% accuracy. This demonstrates a low-cost way to steal intellectual property from additive manufacturing systems using everyday devices.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Command-level accuracy on tested objects may not imply reliable reconstruction for arbitrary G-code sequences","rationale":"The reader's weakest assumption directly identifies the generalization risk; the concrete_test above is a minimal way to check whether that assumption actually holds given the reported numbers.","tokens_in":1730,"tokens_out":277,"duration_ms":16273,"concrete_test":"Partition the collected traces so that all G-code sequences in the test set contain at least one command (or short subsequence) absent from the training set; recompute command-level and full-sequence reconstruction accuracy. If accuracy falls below ~80% on the held-out commands, the emissions do not support the claimed generality.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim of 98.89% command-level reconstruction rests on acoustic/magnetic emissions at 60 cm NLOS containing sufficient unique information. The abstract reports success on 'the final objects' and transfer to a second printer, but provides no detail on whether training/test splits used disjoint G-code sequences, how many distinct commands were involved, or whether accuracy holds when the model encounters previously unseen command combinations or object geometries. If the evaluation re-uses the same limited command vocabulary or object set, high per-command accuracy can be achieved without the signals being informative enough for arbitrary sequences, violating the weakest assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript presents a side-channel attack on additive manufacturing that uses the built-in acoustic and magnetic sensors of two smartphones placed 60 cm away in a non-line-of-sight configuration to capture emissions from a 3D printer. It claims successful reconstruction of the G-code commands of the final printed objects at 98.89% command-level accuracy and reports that the attack strategy transfers to a second printer in a different environment.","tokens_in":1838,"tokens_out":394,"duration_ms":15853,"significance":"If the numerical result and its generalization hold under rigorous validation, the work would establish that consumer-grade smartphones suffice for high-accuracy, non-line-of-sight IP extraction from AM systems, thereby demonstrating a practical and low-cost threat vector that current literature has not addressed with commodity hardware.","major_comments":[{"comment":"Abstract: the central claim of 98.89% command-level reconstruction accuracy is stated without any description of data collection (sampling rates, sensor placement details), feature extraction, classification method, validation procedure (train/test split on G-code sequences or object geometries), or error analysis. This absence makes the numerical result impossible to evaluate against the paper's own evidence and directly undermines assessment of the weakest assumption that the captured emissions contain sufficient unique information for arbitrary G-code sequences.","section":"Abstract"},{"comment":"Abstract: the transferability claim to a second printer is asserted without reporting the accuracy achieved on that printer, whether the model was retrained or transferred zero-shot, the number of distinct commands or objects involved, or controls for environmental differences. These omissions leave the generalization claim unsupported.","section":"Abstract"}],"minor_comments":[],"recommendation":"major_revision","confidential_remarks":"The manuscript appears to lack an experimental-methods section entirely; this is unusual for an empirical security paper and may indicate an incomplete submission rather than a deliberate omission."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback on the abstract. We agree that the abstract would benefit from additional high-level details on methodology and results to better support the claims. We will revise the abstract accordingly while preserving its conciseness. Point-by-point responses follow.","responses":[{"response":"We acknowledge that the abstract, as a concise summary, omits these specifics. The full manuscript details the data collection (smartphone acoustic and magnetic sensors at 60 cm non-line-of-sight), feature extraction, classification approach, validation via train/test splits on G-code sequences and object geometries, and error analysis in the Methods, Experimental Setup, and Results sections, where the 98.89% figure is derived and supported. To address the concern directly, we will revise the abstract to include a brief summary of the sensor modalities, distance/setup, and validation approach.","revision_made":"yes","referee_comment":"[Abstract] Abstract: the central claim of 98.89% command-level reconstruction accuracy is stated without any description of data collection (sampling rates, sensor placement details), feature extraction, classification method, validation procedure (train/test split on G-code sequences or object geometries), or error analysis. This absence makes the numerical result impossible to evaluate against the paper's own evidence and directly undermines assessment of the weakest assumption that the captured emissions contain sufficient unique information for arbitrary G-code sequences."},{"response":"The abstract summarizes the transferability evaluation but does not include the specific accuracy or experimental parameters for the second printer. The manuscript body reports these details (accuracy on the second printer, retraining/transfer procedure, command/object counts, and environmental controls). We will revise the abstract to state the achieved accuracy on the second printer and note the transfer setting.","revision_made":"yes","referee_comment":"[Abstract] Abstract: the transferability claim to a second printer is asserted without reporting the accuracy achieved on that printer, whether the model was retrained or transferred zero-shot, the number of distinct commands or objects involved, or controls for environmental differences. These omissions leave the generalization claim unsupported."}],"tokens_in":1368,"tokens_out":458,"duration_ms":18713,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core contribution is an attack that places two consumer smartphones 60 cm away in non-line-of-sight to capture acoustic and magnetic emissions from a 3D printer and fuse them for G-code reconstruction, plus a transfer test on a second printer. This directly targets the distance, equipment, and line-of-sight limits noted in earlier AM side-channel work.\n\nIt does a clean job of framing the threat model around real-world IP leakage in additive manufacturing and choosing accessible sensors instead of lab gear. The dual-modality choice and the non-LoS distance are concrete steps that make the scenario more plausible than prior single-sensor or close-range attempts.\n\nThe main weakness is the absence of any description of data collection, feature extraction, classifier, train/test split, or error breakdown. The 98.89% command-level figure is stated without evidence that the test objects used disjoint command sequences or geometries from training, so it is impossible to tell whether the signals support arbitrary G-code or only the specific prints evaluated. The transferability result is mentioned but likewise lacks detail on how the environments or printers differed. These gaps make the central accuracy claim unevaluable from the given text.\n\nThe work is aimed at researchers studying side-channel threats to manufacturing systems. A reader already following AM security papers would find the setup description useful as a data point, but the missing experimental specifics limit how much weight the result can carry. It deserves a serious referee to examine whether the full paper supplies the missing validation steps and addresses the arbitrary-sequence concern.","headline":"The paper shows a practical smartphone-based side-channel setup for 3D printer G-code recovery but the 98.89% claim rests on unshown validation details.","tokens_in":2353,"tokens_out":387,"would_cite":false,"duration_ms":15987,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Dual smartphones reconstruct 3D printer G-code commands from acoustic and magnetic emissions at 60 cm distance.","keywords":["side-channel attack","additive manufacturing","3D printing","IP theft","G-code reconstruction","acoustic emissions","magnetic emissions","smartphone sensors"],"falsifier":"Reconstruction accuracy falling below 80 percent when the same method is applied to a third printer model or a substantially altered room layout would show the claimed transferability does not hold.","tokens_in":2634,"feed_emoji":"📱","tokens_out":608,"duration_ms":16380,"temperature":0.7,"pith_summary":"The paper shows that two ordinary smartphones placed 60 cm away can capture acoustic and magnetic signals from a 3D printer even without line of sight. These signals are processed to recover the exact sequence of G-code commands that define the printed object, reaching 98.89 percent accuracy at the individual command level. The same approach transfers to a second printer in a different setting. Because G-code directly encodes the intellectual property of the manufactured part, the result demonstrates a practical route for unauthorized extraction of design data using only consumer devices.","feed_headline":"Smartphones reconstruct 3D printer G-code with 98.89% accuracy","feed_subtitle":"Dual phones capture acoustic and magnetic signals from 60 cm away in non-line-of-sight to recover printing commands.","key_machinery":"Multi-modality side-channel using smartphone acoustic and magnetic sensors to reconstruct G-code sequences.","core_discovery":"The attack uses dual smartphones' internal sensors to collect acoustic and magnetic emissions from a 3D printer at 60 cm in non-line-of-sight setup. It reconstructs the G-code commands of the final objects at 98.89% command-level reconstruction accuracy and demonstrates transferability to another printer in a different environment.","pith_inferences":["Similar emissions from other computer-controlled manufacturing tools could be targeted with the same dual-sensor approach.","Defenses would need to address both acoustic and magnetic leakage simultaneously rather than one channel alone.","Extending the attack to additional sensor modalities on the same phones might further raise reconstruction rates."],"forward_implications":["G-code reconstruction at 98.89 percent accuracy directly exposes the intellectual property of printed objects.","The attack succeeds at 60 cm distance without requiring line of sight or specialized hardware.","Transferability across printers and environments indicates the method is not limited to one specific machine or location.","The results establish that consumer smartphones alone are sufficient to perform this side-channel extraction."],"fun_headline_variants":["Dual smartphones reconstruct 3D printer G-code 98.89% accurately at 60 cm","98.89% command reconstruction using dual phones from 60 cm non-line-of-sight","Two phones capture emissions to achieve 98.89% G-code reconstruction","Non-line-of-sight dual smartphones achieve 98.89% accurate G-code recovery"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The acoustic and magnetic emissions captured by smartphone sensors at 60 cm contain sufficient unique and consistent information to enable high-accuracy reconstruction of arbitrary G-code sequences across different printers and environments.","fun_headline_variants_meta":{"raw":{"variants":["Dual smartphones reconstruct 3D printer G-code 98.89% accurately at 60 cm","98.89% command reconstruction using dual phones from 60 cm non-line-of-sight","Two phones capture emissions to achieve 98.89% G-code reconstruction","Non-line-of-sight dual smartphones achieve 98.89% accurate G-code recovery"]},"model":"grok-4.3","cost_usd":0.008387,"raw_usage":{"total_tokens":3798,"prompt_tokens":671,"num_sources_used":0,"completion_tokens":89,"cost_in_usd_ticks":83874500,"prompt_tokens_details":{"text_tokens":671,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3038,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":671,"tokens_out":89,"duration_ms":28197,"temperature":1.0,"reasoning_tokens":3038,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-02T18:39:21.860816+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Reconstruction accuracy falling below 80 percent when the same method is applied to a third printer model or a substantially altered room layout would show the claimed transferability does not hold.","supporting_citations":[],"review_version":1}