{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:CQ445C746WIGW6BDTQXYHMWICT","short_pith_number":"pith:CQ445C74","canonical_record":{"source":{"id":"2605.21706","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-20T20:10:27Z","cross_cats_sorted":[],"title_canon_sha256":"5ffe54f8167887b5380ff2946232931878e0c01e7bb655df66cb3d79f3040ca0","abstract_canon_sha256":"b9ddfd7c9e4f5e1dd085afb5343cece6294dc555a026caf2de8fd407b6fd3589"},"schema_version":"1.0"},"canonical_sha256":"1439ce8bfcf5906b78239c2f83b2c814f8dc44852a974542099c38c797f0764a","source":{"kind":"arxiv","id":"2605.21706","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.21706","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"arxiv_version","alias_value":"2605.21706v1","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.21706","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"pith_short_12","alias_value":"CQ445C746WIG","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"pith_short_16","alias_value":"CQ445C746WIGW6BD","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"pith_short_8","alias_value":"CQ445C74","created_at":"2026-05-22T01:03:28Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:CQ445C746WIGW6BDTQXYHMWICT","target":"record","payload":{"canonical_record":{"source":{"id":"2605.21706","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-20T20:10:27Z","cross_cats_sorted":[],"title_canon_sha256":"5ffe54f8167887b5380ff2946232931878e0c01e7bb655df66cb3d79f3040ca0","abstract_canon_sha256":"b9ddfd7c9e4f5e1dd085afb5343cece6294dc555a026caf2de8fd407b6fd3589"},"schema_version":"1.0"},"canonical_sha256":"1439ce8bfcf5906b78239c2f83b2c814f8dc44852a974542099c38c797f0764a","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T01:03:28.799856Z","signature_b64":"Lxlr0UDqBJqWWaUBj53Vt2G9C9AHMEvwp1czh+qAanN6a7ylUhTmvpQ493ub2TGNbt2ZuwAoYBrNpvT97fibAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1439ce8bfcf5906b78239c2f83b2c814f8dc44852a974542099c38c797f0764a","last_reissued_at":"2026-05-22T01:03:28.799371Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T01:03:28.799371Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.21706","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-22T01:03:28Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"lEiSsMaNR+NBPR3989BHoieibUnfwNL5UaqzlmOUm2Wcmgjd04hKGeICSegFmdZwpVv5NoKE8IQFx/6dcjsKAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T11:18:44.396807Z"},"content_sha256":"389e703afa71364989e8dd272d33dd683f1feee075b75399712d8c96d86faea9","schema_version":"1.0","event_id":"sha256:389e703afa71364989e8dd272d33dd683f1feee075b75399712d8c96d86faea9"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:CQ445C746WIGW6BDTQXYHMWICT","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Latent-space Attacks for Refusal Evasion in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Battista Biggio, Fabio Brau, Fabio Roli, Giorgio Piras, Luca Oneto, Maura Pintor, Raffaele Mura","submitted_at":"2026-05-20T20:10:27Z","abstract_excerpt":"Safety-aligned language models are trained to refuse harmful requests, yet refusal behavior can be suppressed by steering their internal representations. Existing methods do so by ablating a refusal direction from model activations, aiming to remove refusal from the model's residual stream. Despite their empirical success, these methods lack a principled account of the latent-space transformation they induce and why it suppresses refusal. In this work, we recast refusal suppression as a latent-space evasion attack against linear probes trained to separate refused from answered prompts. Under t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.21706","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.21706/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-22T01:03:28Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"I7gwXAC2qeRhLQSWYs6yRn/toNXvyJNjCQrdPbwZGDm3LMPl8C4+dbqOh849tuMwEgQ0yMfjyrJydNJ97iMGDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T11:18:44.397507Z"},"content_sha256":"934378555b23c6dffbe98bbac9ed46e64780ea4d346e6605157a353c15bcdab7","schema_version":"1.0","event_id":"sha256:934378555b23c6dffbe98bbac9ed46e64780ea4d346e6605157a353c15bcdab7"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/CQ445C746WIGW6BDTQXYHMWICT/bundle.json","state_url":"https://pith.science/pith/CQ445C746WIGW6BDTQXYHMWICT/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/CQ445C746WIGW6BDTQXYHMWICT/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-27T11:18:44Z","links":{"resolver":"https://pith.science/pith/CQ445C746WIGW6BDTQXYHMWICT","bundle":"https://pith.science/pith/CQ445C746WIGW6BDTQXYHMWICT/bundle.json","state":"https://pith.science/pith/CQ445C746WIGW6BDTQXYHMWICT/state.json","well_known_bundle":"https://pith.science/.well-known/pith/CQ445C746WIGW6BDTQXYHMWICT/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:CQ445C746WIGW6BDTQXYHMWICT","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":"b9ddfd7c9e4f5e1dd085afb5343cece6294dc555a026caf2de8fd407b6fd3589","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-20T20:10:27Z","title_canon_sha256":"5ffe54f8167887b5380ff2946232931878e0c01e7bb655df66cb3d79f3040ca0"},"schema_version":"1.0","source":{"id":"2605.21706","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.21706","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"arxiv_version","alias_value":"2605.21706v1","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.21706","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"pith_short_12","alias_value":"CQ445C746WIG","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"pith_short_16","alias_value":"CQ445C746WIGW6BD","created_at":"2026-05-22T01:03:28Z"},{"alias_kind":"pith_short_8","alias_value":"CQ445C74","created_at":"2026-05-22T01:03:28Z"}],"graph_snapshots":[{"event_id":"sha256:934378555b23c6dffbe98bbac9ed46e64780ea4d346e6605157a353c15bcdab7","target":"graph","created_at":"2026-05-22T01:03:28Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2605.21706/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Safety-aligned language models are trained to refuse harmful requests, yet refusal behavior can be suppressed by steering their internal representations. Existing methods do so by ablating a refusal direction from model activations, aiming to remove refusal from the model's residual stream. Despite their empirical success, these methods lack a principled account of the latent-space transformation they induce and why it suppresses refusal. In this work, we recast refusal suppression as a latent-space evasion attack against linear probes trained to separate refused from answered prompts. Under t","authors_text":"Battista Biggio, Fabio Brau, Fabio Roli, Giorgio Piras, Luca Oneto, Maura Pintor, Raffaele Mura","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-20T20:10:27Z","title":"Latent-space Attacks for Refusal Evasion in Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.21706","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:389e703afa71364989e8dd272d33dd683f1feee075b75399712d8c96d86faea9","target":"record","created_at":"2026-05-22T01:03:28Z","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":"b9ddfd7c9e4f5e1dd085afb5343cece6294dc555a026caf2de8fd407b6fd3589","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-20T20:10:27Z","title_canon_sha256":"5ffe54f8167887b5380ff2946232931878e0c01e7bb655df66cb3d79f3040ca0"},"schema_version":"1.0","source":{"id":"2605.21706","kind":"arxiv","version":1}},"canonical_sha256":"1439ce8bfcf5906b78239c2f83b2c814f8dc44852a974542099c38c797f0764a","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"1439ce8bfcf5906b78239c2f83b2c814f8dc44852a974542099c38c797f0764a","first_computed_at":"2026-05-22T01:03:28.799371Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-22T01:03:28.799371Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Lxlr0UDqBJqWWaUBj53Vt2G9C9AHMEvwp1czh+qAanN6a7ylUhTmvpQ493ub2TGNbt2ZuwAoYBrNpvT97fibAw==","signature_status":"signed_v1","signed_at":"2026-05-22T01:03:28.799856Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.21706","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:389e703afa71364989e8dd272d33dd683f1feee075b75399712d8c96d86faea9","sha256:934378555b23c6dffbe98bbac9ed46e64780ea4d346e6605157a353c15bcdab7"],"state_sha256":"a391e82408eb32c00640ec13b42c96544fba08a7361bb439750994666ecf9cf6"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"UFWaXE4VB7J+a9OUYr/Hw28X2HuLpWmtWI6aNeW88jH1QOnNhofOe4b7am+N/WwEY+upQzMZ+qeaRyzsOJ1GAg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-27T11:18:44.401291Z","bundle_sha256":"f8cfcf5e5d981c9fae594ef5fafd5b0975525933588b543d03224c322466b08a"}}