{"id":"4a2c4d1b8e30d6d614dfb1667a15834bbae447856b22cba6f915b988cabc0f7d","title":"arXiv cs.AI paper: ConfGuard checks for predictive-maintenance agents","argument":"The authors of “ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents” argue that industrial agents need checks of whether predictions fit an asset’s degradation physics, not only checks of data provenance or tool access. ConfGuard’s conformal monitor tests predictions against the asset’s history: capped remaining life must not rise, and may fall by at most one cycle per cycle. A second validator compares the agent’s maintenance-urgency decision with a deterministic policy applied to the same interval. On 10,196 C-MAPSS test windows, the authors report that the monitor reduced a covert attacker’s reach on healthy assets from 0.529 to 0.046 and closed suppression of failing assets from 0.151 to 0. It escalated 0.027 of clean windows against a 0.05 target. With a Qwen3-14B planner, the monitors held urgency-class injection to 0.00 and a universal sensor attack to 0.12, versus 0.29 and 0.54 for a defended prompt. These reported test results do not establish performance in deployed industrial systems. The authors also report an attack class that the projection misses, but the abstract does not identify it.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results are for C-MAPSS test windows; the supplied abstract does not establish how the monitors perform in deployed industrial systems. The abstract says the projection misses an attack class but does not identify it or describe its effect.","speaker":{"name":"V, Arunkumar, R., Gangadharan G., Senthilkumar, S.","roleAtSource":"Article author"},"source":{"title":"ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.22288","publishedAt":"2026-11-17","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that existing defenses focused on data provenance and tool privilege do not check whether a prognostic quantity is physically plausible.","excerpt":"Recent agent defenses police data provenance and tool privilege but cannot tell whether a prognostic quantity is physically plausible.","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"},{"text":"ConfGuard’s conformal monitor checks predictions against degradation physics and the asset's admitted history, requiring capped remaining life not to rise and to fall at most one cycle per cycle.","excerpt":"Its conformal monitor tests each prediction, anchored to the asset's admitted history, against degradation physics: capped remaining life never rises and falls at most one cycle per cycle.","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"},{"text":"The paper reports that on 10,196 C-MAPSS test windows, the monitor reduced covert-attack reach on healthy assets from 0.529 to 0.046 and closed suppression of failing assets from 0.151 to 0, while escalating 0.027 of clean windows against a 0.05 target.","excerpt":"On all 10,196 C-MAPSS test windows the monitor cuts a covert attacker's reach on healthy assets from 0.529 to 0.046 and closes suppression of failing assets (0.151 to 0), escalating 0.027 of clean windows against a 0.05 target","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"},{"text":"With a Qwen3-14B planner, the authors report lower attack results for composed monitors than for a defended prompt, for urgency-class injection and a universal sensor attack.","excerpt":"Behind a Qwen3-14B planner the composed monitors hold urgency-class injection to 0.00 and the universal sensor attack to 0.12, against 0.29 and 0.54 for a defended prompt.","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"},{"text":"The abstract reports that the projection misses an attack class but does not identify that class.","excerpt":"We report an attack class the projection misses.","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22288","checkedAt":"2026-11-18T15:50:09.964Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What attack class does the projection miss, and how does it affect the reported protection?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI paper: ConfGuard checks for predictive-maintenance agents","summary":"The authors of “ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents” argue that industrial agents need checks of whether predictions fit an asset’s degradation physics, not only checks of data provenance or tool access. ConfGuard’s conformal monitor tests predictions against the asset’s history: capped remaining life must not rise, and may fall by at most one cycle per cycle. A second validator compares the agent’s maintenance-urgency decision with a deterministic policy applied to the same interval. On 10,196 C-MAPSS test windows, the authors report that the monitor reduced a covert attacker’s reach on healthy assets from 0.529 to 0.046 and closed suppression of failing assets from 0.151 to 0. It escalated 0.027 of clean windows against a 0.05 target. With a Qwen3-14B planner, the monitors held urgency-class injection to 0.00 and a universal sensor attack to 0.12, versus 0.29 and 0.54 for a defended prompt. These reported test results do not establish performance in deployed industrial systems. The authors also report an attack class that the projection misses, but the abstract does not identify it.","question":"What attack class does the projection miss, and how does it affect the reported protection?","claims":[{"text":"The authors argue that existing defenses focused on data provenance and tool privilege do not check whether a prognostic quantity is physically plausible.","excerpt":"Recent agent defenses police data provenance and tool privilege but cannot tell whether a prognostic quantity is physically plausible.","support":"source-stated"},{"text":"ConfGuard’s conformal monitor checks predictions against degradation physics and the asset's admitted history, requiring capped remaining life not to rise and to fall at most one cycle per cycle.","excerpt":"Its conformal monitor tests each prediction, anchored to the asset's admitted history, against degradation physics: capped remaining life never rises and falls at most one cycle per cycle.","support":"source-stated"},{"text":"The paper reports that on 10,196 C-MAPSS test windows, the monitor reduced covert-attack reach on healthy assets from 0.529 to 0.046 and closed suppression of failing assets from 0.151 to 0, while escalating 0.027 of clean windows against a 0.05 target.","excerpt":"On all 10,196 C-MAPSS test windows the monitor cuts a covert attacker's reach on healthy assets from 0.529 to 0.046 and closes suppression of failing assets (0.151 to 0), escalating 0.027 of clean windows against a 0.05 target","support":"source-stated"},{"text":"With a Qwen3-14B planner, the authors report lower attack results for composed monitors than for a defended prompt, for urgency-class injection and a universal sensor attack.","excerpt":"Behind a Qwen3-14B planner the composed monitors hold urgency-class injection to 0.00 and the universal sensor attack to 0.12, against 0.29 and 0.54 for a defended prompt.","support":"source-stated"},{"text":"The abstract reports that the projection misses an attack class but does not identify that class.","excerpt":"We report an attack class the projection misses.","support":"source-stated"}],"limitations":["The reported results are for C-MAPSS test windows; the supplied abstract does not establish how the monitors perform in deployed industrial systems.","The abstract says the projection misses an attack class but does not identify it or describe its effect."],"mediaId":"f3f5e8f906dec8fe47b20d6e0cb93acfbe14bdacda968baa7fa7bfc457d79539","correctionReason":null,"id":"4a2c4d1b8e30d6d614dfb1667a15834bbae447856b22cba6f915b988cabc0f7d","publicationId":"45075be1-d672-415d-85cd-df28e5ba528f","version":1,"sourceUrl":"https://arxiv.org/abs/2611.22288","sourceTitle":"ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-17","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-18T05:07:16.093Z","createdAt":"2026-11-18T15:19:18.908Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"f3f5e8f906dec8fe47b20d6e0cb93acfbe14bdacda968baa7fa7bfc457d79539","kind":"article","url":"https://arxiv.org/abs/2611.22288","sourceUrl":"https://arxiv.org/abs/2611.22288","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-18T15:18:23.383Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22288","checkedAt":"2026-11-18T15:50:09.964Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"f3f5e8f906dec8fe47b20d6e0cb93acfbe14bdacda968baa7fa7bfc457d79539","confidence":0.53,"provider":"typesafe","model":"jev-1.13.0","inputHash":"e7c6ae80999029c8a6ef0e5ebe912510160e1cb160c46e34e8b2491558edd43d","decidedAt":"2026-11-18T15:18:24.519Z"},"provenance":{"textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observationId":"648f09874481b98c94631f542e6f514bc5662e2a8afbe60aa059d24f956f060f","retrievedAt":"2026-11-18T15:18:23.168Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"0bb53a14-ad90-47ef-804a-bf22944b1473"},"reviewer":{"model":"gpt-6-luna","runId":"ea287aca-b21e-42fb-8ffe-0ab7fd378955"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-18T15:19:18.908Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"V, Arunkumar","url":null,"kind":"unknown"},{"name":"R., Gangadharan G.","url":null,"kind":"unknown"},{"name":"Senthilkumar, S.","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"V, Arunkumar","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"V, Arunkumar","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22288","locator":"head/meta[name=author|citation_author]"},"assertionId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T15:50:09.964Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d","sourceUrl":"https://arxiv.org/abs/2611.22288","contentSha256":"e434144c9e497e2800786bfb26a5e66892f74acaf6aa37b4500a56f90ec3008b","textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observedAt":"2026-11-18T15:50:09.964Z"}},{"actor":{"name":"R., Gangadharan G.","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"R., Gangadharan G.","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22288","locator":"head/meta[name=author|citation_author]"},"assertionId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T15:50:09.964Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d","sourceUrl":"https://arxiv.org/abs/2611.22288","contentSha256":"e434144c9e497e2800786bfb26a5e66892f74acaf6aa37b4500a56f90ec3008b","textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observedAt":"2026-11-18T15:50:09.964Z"}},{"actor":{"name":"Senthilkumar, S.","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Senthilkumar, S.","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22288","locator":"head/meta[name=author|citation_author]"},"assertionId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T15:50:09.964Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d","sourceUrl":"https://arxiv.org/abs/2611.22288","contentSha256":"e434144c9e497e2800786bfb26a5e66892f74acaf6aa37b4500a56f90ec3008b","textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observedAt":"2026-11-18T15:50:09.964Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.22288","snapshotId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d","contentSha256":"e434144c9e497e2800786bfb26a5e66892f74acaf6aa37b4500a56f90ec3008b","textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observedAt":"2026-11-18T15:50:09.964Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/4a2c4d1b8e30d6d614dfb1667a15834bbae447856b22cba6f915b988cabc0f7d/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-paper-confguard-checks-for-predictive-maintenance-4a2c4d1b","formats":{"html":"/readings/arxiv-cs-ai-paper-confguard-checks-for-predictive-maintenance-4a2c4d1b","markdown":"/readings/arxiv-cs-ai-paper-confguard-checks-for-predictive-maintenance-4a2c4d1b.md","json":"/readings/arxiv-cs-ai-paper-confguard-checks-for-predictive-maintenance-4a2c4d1b.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"conditions":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-llm","name":"Large language model","explanation":"A language model with many learned numerical settings, used to process and generate language.","referenceUrl":"https://developers.google.com/machine-learning/glossary#large-language-model"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"4a2c4d1b8e30d6d614dfb1667a15834bbae447856b22cba6f915b988cabc0f7d","revision":"45075be1-d672-415d-85cd-df28e5ba528f","revisionKind":"publication-id","title":"arXiv cs.AI paper: ConfGuard checks for predictive-maintenance agents","url":"https://escapemeridian.com/readings/arxiv-cs-ai-paper-confguard-checks-for-predictive-maintenance-4a2c4d1b"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.22288","revision":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","revisionKind":"sha256-source-text","title":"ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents","url":"https://arxiv.org/abs/2611.22288"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter."},"revisions":[{"id":"4a2c4d1b8e30d6d614dfb1667a15834bbae447856b22cba6f915b988cabc0f7d","title":"arXiv cs.AI paper: ConfGuard checks for predictive-maintenance agents","argument":"The authors of “ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents” argue that industrial agents need checks of whether predictions fit an asset’s degradation physics, not only checks of data provenance or tool access. ConfGuard’s conformal monitor tests predictions against the asset’s history: capped remaining life must not rise, and may fall by at most one cycle per cycle. A second validator compares the agent’s maintenance-urgency decision with a deterministic policy applied to the same interval. On 10,196 C-MAPSS test windows, the authors report that the monitor reduced a covert attacker’s reach on healthy assets from 0.529 to 0.046 and closed suppression of failing assets from 0.151 to 0. It escalated 0.027 of clean windows against a 0.05 target. With a Qwen3-14B planner, the monitors held urgency-class injection to 0.00 and a universal sensor attack to 0.12, versus 0.29 and 0.54 for a defended prompt. These reported test results do not establish performance in deployed industrial systems. The authors also report an attack class that the projection misses, but the abstract does not identify it.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results are for C-MAPSS test windows; the supplied abstract does not establish how the monitors perform in deployed industrial systems. The abstract says the projection misses an attack class but does not identify it or describe its effect.","speaker":{"name":"V, Arunkumar, R., Gangadharan G., Senthilkumar, S.","roleAtSource":"Article author"},"source":{"title":"ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.22288","publishedAt":"2026-11-17","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that existing defenses focused on data provenance and tool privilege do not check whether a prognostic quantity is physically plausible.","excerpt":"Recent agent defenses police data provenance and tool privilege but cannot tell whether a prognostic quantity is physically plausible.","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"},{"text":"ConfGuard’s conformal monitor checks predictions against degradation physics and the asset's admitted history, requiring capped remaining life not to rise and to fall at most one cycle per cycle.","excerpt":"Its conformal monitor tests each prediction, anchored to the asset's admitted history, against degradation physics: capped remaining life never rises and falls at most one cycle per cycle.","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"},{"text":"The paper reports that on 10,196 C-MAPSS test windows, the monitor reduced covert-attack reach on healthy assets from 0.529 to 0.046 and closed suppression of failing assets from 0.151 to 0, while escalating 0.027 of clean windows against a 0.05 target.","excerpt":"On all 10,196 C-MAPSS test windows the monitor cuts a covert attacker's reach on healthy assets from 0.529 to 0.046 and closes suppression of failing assets (0.151 to 0), escalating 0.027 of clean windows against a 0.05 target","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"},{"text":"With a Qwen3-14B planner, the authors report lower attack results for composed monitors than for a defended prompt, for urgency-class injection and a universal sensor attack.","excerpt":"Behind a Qwen3-14B planner the composed monitors hold urgency-class injection to 0.00 and the universal sensor attack to 0.12, against 0.29 and 0.54 for a defended prompt.","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"},{"text":"The abstract reports that the projection misses an attack class but does not identify that class.","excerpt":"We report an attack class the projection misses.","url":"https://arxiv.org/abs/2611.22288","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22288","checkedAt":"2026-11-18T15:50:09.964Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What attack class does the projection miss, and how does it affect the reported protection?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI paper: ConfGuard checks for predictive-maintenance agents","summary":"The authors of “ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents” argue that industrial agents need checks of whether predictions fit an asset’s degradation physics, not only checks of data provenance or tool access. ConfGuard’s conformal monitor tests predictions against the asset’s history: capped remaining life must not rise, and may fall by at most one cycle per cycle. A second validator compares the agent’s maintenance-urgency decision with a deterministic policy applied to the same interval. On 10,196 C-MAPSS test windows, the authors report that the monitor reduced a covert attacker’s reach on healthy assets from 0.529 to 0.046 and closed suppression of failing assets from 0.151 to 0. It escalated 0.027 of clean windows against a 0.05 target. With a Qwen3-14B planner, the monitors held urgency-class injection to 0.00 and a universal sensor attack to 0.12, versus 0.29 and 0.54 for a defended prompt. These reported test results do not establish performance in deployed industrial systems. The authors also report an attack class that the projection misses, but the abstract does not identify it.","question":"What attack class does the projection miss, and how does it affect the reported protection?","claims":[{"text":"The authors argue that existing defenses focused on data provenance and tool privilege do not check whether a prognostic quantity is physically plausible.","excerpt":"Recent agent defenses police data provenance and tool privilege but cannot tell whether a prognostic quantity is physically plausible.","support":"source-stated"},{"text":"ConfGuard’s conformal monitor checks predictions against degradation physics and the asset's admitted history, requiring capped remaining life not to rise and to fall at most one cycle per cycle.","excerpt":"Its conformal monitor tests each prediction, anchored to the asset's admitted history, against degradation physics: capped remaining life never rises and falls at most one cycle per cycle.","support":"source-stated"},{"text":"The paper reports that on 10,196 C-MAPSS test windows, the monitor reduced covert-attack reach on healthy assets from 0.529 to 0.046 and closed suppression of failing assets from 0.151 to 0, while escalating 0.027 of clean windows against a 0.05 target.","excerpt":"On all 10,196 C-MAPSS test windows the monitor cuts a covert attacker's reach on healthy assets from 0.529 to 0.046 and closes suppression of failing assets (0.151 to 0), escalating 0.027 of clean windows against a 0.05 target","support":"source-stated"},{"text":"With a Qwen3-14B planner, the authors report lower attack results for composed monitors than for a defended prompt, for urgency-class injection and a universal sensor attack.","excerpt":"Behind a Qwen3-14B planner the composed monitors hold urgency-class injection to 0.00 and the universal sensor attack to 0.12, against 0.29 and 0.54 for a defended prompt.","support":"source-stated"},{"text":"The abstract reports that the projection misses an attack class but does not identify that class.","excerpt":"We report an attack class the projection misses.","support":"source-stated"}],"limitations":["The reported results are for C-MAPSS test windows; the supplied abstract does not establish how the monitors perform in deployed industrial systems.","The abstract says the projection misses an attack class but does not identify it or describe its effect."],"mediaId":"f3f5e8f906dec8fe47b20d6e0cb93acfbe14bdacda968baa7fa7bfc457d79539","correctionReason":null,"id":"4a2c4d1b8e30d6d614dfb1667a15834bbae447856b22cba6f915b988cabc0f7d","publicationId":"45075be1-d672-415d-85cd-df28e5ba528f","version":1,"sourceUrl":"https://arxiv.org/abs/2611.22288","sourceTitle":"ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-17","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-18T05:07:16.093Z","createdAt":"2026-11-18T15:19:18.908Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"f3f5e8f906dec8fe47b20d6e0cb93acfbe14bdacda968baa7fa7bfc457d79539","kind":"article","url":"https://arxiv.org/abs/2611.22288","sourceUrl":"https://arxiv.org/abs/2611.22288","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-18T15:18:23.383Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22288","checkedAt":"2026-11-18T15:50:09.964Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"f3f5e8f906dec8fe47b20d6e0cb93acfbe14bdacda968baa7fa7bfc457d79539","confidence":0.53,"provider":"typesafe","model":"jev-1.13.0","inputHash":"e7c6ae80999029c8a6ef0e5ebe912510160e1cb160c46e34e8b2491558edd43d","decidedAt":"2026-11-18T15:18:24.519Z"},"provenance":{"textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observationId":"648f09874481b98c94631f542e6f514bc5662e2a8afbe60aa059d24f956f060f","retrievedAt":"2026-11-18T15:18:23.168Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"0bb53a14-ad90-47ef-804a-bf22944b1473"},"reviewer":{"model":"gpt-6-luna","runId":"ea287aca-b21e-42fb-8ffe-0ab7fd378955"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-18T15:19:18.908Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"V, Arunkumar","url":null,"kind":"unknown"},{"name":"R., Gangadharan G.","url":null,"kind":"unknown"},{"name":"Senthilkumar, S.","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"V, Arunkumar","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"V, Arunkumar","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22288","locator":"head/meta[name=author|citation_author]"},"assertionId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T15:50:09.964Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d","sourceUrl":"https://arxiv.org/abs/2611.22288","contentSha256":"e434144c9e497e2800786bfb26a5e66892f74acaf6aa37b4500a56f90ec3008b","textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observedAt":"2026-11-18T15:50:09.964Z"}},{"actor":{"name":"R., Gangadharan G.","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"R., Gangadharan G.","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22288","locator":"head/meta[name=author|citation_author]"},"assertionId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T15:50:09.964Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d","sourceUrl":"https://arxiv.org/abs/2611.22288","contentSha256":"e434144c9e497e2800786bfb26a5e66892f74acaf6aa37b4500a56f90ec3008b","textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observedAt":"2026-11-18T15:50:09.964Z"}},{"actor":{"name":"Senthilkumar, S.","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Senthilkumar, S.","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22288","locator":"head/meta[name=author|citation_author]"},"assertionId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T15:50:09.964Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d","sourceUrl":"https://arxiv.org/abs/2611.22288","contentSha256":"e434144c9e497e2800786bfb26a5e66892f74acaf6aa37b4500a56f90ec3008b","textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observedAt":"2026-11-18T15:50:09.964Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.22288","snapshotId":"d8a4960be73ba421f277aabd2b78c9245983a0ca138532fc7328b6f09de9940d","contentSha256":"e434144c9e497e2800786bfb26a5e66892f74acaf6aa37b4500a56f90ec3008b","textHash":"776ee9a64a543e7a1ead34e1c93a90830db50467e74c4cd8ceca1b27e032a862","observedAt":"2026-11-18T15:50:09.964Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/4a2c4d1b8e30d6d614dfb1667a15834bbae447856b22cba6f915b988cabc0f7d/source"}}}],"audits":[{"publicationId":"45075be1-d672-415d-85cd-df28e5ba528f","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the arguments and results to the authors of the named arXiv paper. It presents the reported numbers as author-reported results, not established facts.","excerpt":"Authors:Arunkumar V, Gangadharan G. R., S. Senthilkumar"},{"dimension":"summary","status":"uncertain","explanation":"The figures are accurate, but a non-specialist would struggle to follow the reading. It never explains what these agents do: read sensor windows, retrieve procedures and dispatch work orders. It also omits the two attack channels the monitor defends against: injected text in retrieved documents, and bounded sensor perturbations that corrupt remaining-useful-life estimates. Without these, 'covert attacker', 'suppression', 'urgency-class injection' and 'capped remaining life' are hard to read. 'Projection' is also unexplained. Smallest repair: add one sentence describing the agent's loop and the two attack channels. It would also help to note that the abstract says a fixed physical range leaves both attacks open.","excerpt":"Two attack channels meet there: injected text in a retrieved document can hijack the decision, and a bounded sensor perturbation can corrupt the remaining-useful-life estimate behind it."},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-11-17 matches the v1 submission date on the arXiv page.","excerpt":"[Submitted on 17 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim matches the abstract's wording and numbers. The limitation stating the results do not establish deployed performance is appropriately cautious. The question about the unidentified attack class follows from the abstract and does not assume an answer.","excerpt":"We report an attack class the projection misses."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the original arXiv abstract page. This is the most direct way for readers to access the source.","excerpt":null}],"createdAt":"2026-11-18T15:19:18.908Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}