{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:45C3OQ32U6Q5XDEWZO4GKM3AXZ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"c8b0f935c930efed2220208d4f5886b601e689720c7b0ebbb4674ba75354df7b","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-11T21:48:03Z","title_canon_sha256":"b19d9a16ee1e13b496c430be41e2c32586c75a70fac747d5c292d330614277e5"},"schema_version":"1.0","source":{"id":"2603.11331","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.11331","created_at":"2026-05-29T02:05:42Z"},{"alias_kind":"arxiv_version","alias_value":"2603.11331v3","created_at":"2026-05-29T02:05:42Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.11331","created_at":"2026-05-29T02:05:42Z"},{"alias_kind":"pith_short_12","alias_value":"45C3OQ32U6Q5","created_at":"2026-05-29T02:05:42Z"},{"alias_kind":"pith_short_16","alias_value":"45C3OQ32U6Q5XDEW","created_at":"2026-05-29T02:05:42Z"},{"alias_kind":"pith_short_8","alias_value":"45C3OQ32","created_at":"2026-05-29T02:05:42Z"}],"graph_snapshots":[{"event_id":"sha256:aaabebb4213cebe089bb367c22dbf2623dc2bcccefc881e4f28c1a19fee2c25e","target":"graph","created_at":"2026-05-29T02:05:42Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"strong adversarial prompt-injection attacks can amplify attack success rate from the slow polynomial growth observed without injection to exponential growth with the number of inference-time samples"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"a small set of assumptions on the distribution of safe generation across contexts under which both scaling laws follow"},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Strong adversarial prompt injections shift jailbreak success scaling from polynomial to exponential with more inference samples, derived from a spin-glass generative model of LLM outputs."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Strong adversarial prompt injections turn slow polynomial growth of jailbreak success into exponential growth with more inference samples."}],"snapshot_sha256":"62c11c0d1a70bcc0f3c9f75dcc2ee35722dead7734d613a721ab458319785aca"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"1fe285041d1e5885d3400c8cde56fc8851d3c4936f4553eea7c537da583f13f4"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2603.11331/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Adversarial attacks can reliably steer safety-aligned large language models toward unsafe behavior. Empirically, we find that adversarial prompt-injection attacks can amplify attack success rate from the slow polynomial growth observed without injection to exponential growth with the number of inference-time samples. We first identify a minimal statistical mechanism for these two regimes by giving a small set of assumptions on the distribution of safe generation across contexts under which both scaling laws follow. To explain this phenomenon further, we propose a theoretical generative model o","authors_text":"Annesya Banerjee, Cengiz Pehlevan, Indranil Halder","cross_cats":["cs.AI"],"headline":"Strong adversarial prompt injections turn slow polynomial growth of jailbreak success into exponential growth with more inference samples.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-11T21:48:03Z","title":"Jailbreak Scaling Laws for Large Language Models: Polynomial-Exponential Crossover"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.11331","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T12:37:31.020882Z","id":"46c5e45c-d2dd-4f7c-baf8-c678a02e2eb5","model_set":{"reader":"grok-4.3"},"one_line_summary":"Strong adversarial prompt injections shift jailbreak success scaling from polynomial to exponential with more inference samples, derived from a spin-glass generative model of LLM outputs.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Strong adversarial prompt injections turn slow polynomial growth of jailbreak success into exponential growth with more inference samples.","strongest_claim":"strong adversarial prompt-injection attacks can amplify attack success rate from the slow polynomial growth observed without injection to exponential growth with the number of inference-time samples","weakest_assumption":"a small set of assumptions on the distribution of safe generation across contexts under which both scaling laws follow"}},"verdict_id":"46c5e45c-d2dd-4f7c-baf8-c678a02e2eb5"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:99caad75b02acb55d976045c21fc13e47d097a4b4546cb9edcfeb110d8f5422c","target":"record","created_at":"2026-05-29T02:05:42Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"c8b0f935c930efed2220208d4f5886b601e689720c7b0ebbb4674ba75354df7b","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-11T21:48:03Z","title_canon_sha256":"b19d9a16ee1e13b496c430be41e2c32586c75a70fac747d5c292d330614277e5"},"schema_version":"1.0","source":{"id":"2603.11331","kind":"arxiv","version":3}},"canonical_sha256":"e745b7437aa7a1db8c96cbb8653360be45e3a9d57395fceac1eb82902d85898d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"e745b7437aa7a1db8c96cbb8653360be45e3a9d57395fceac1eb82902d85898d","first_computed_at":"2026-05-29T02:05:42.783753Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-29T02:05:42.783753Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"a5wGjQP+HNJ83WoqnOfJUgC8B/mO9JJ6ISZwADC+BMrB265obXYBwevKEIlViJWzht10yKceHnyDpoxO/isjCg==","signature_status":"signed_v1","signed_at":"2026-05-29T02:05:42.784457Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.11331","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:99caad75b02acb55d976045c21fc13e47d097a4b4546cb9edcfeb110d8f5422c","sha256:aaabebb4213cebe089bb367c22dbf2623dc2bcccefc881e4f28c1a19fee2c25e"],"state_sha256":"833b47076f70a0058a392b557ede87dfd0b5cda9a383dc3d70bfbe08c300cf6e"}