{"_canonicalization":{"envelope_id":"axm_ + sha256(envelope minus {signature, axiom_id, anchors})","envelope_signature":"ed25519(envelope minus {signature, axiom_id})","json":"sort_keys=True, separators=(',',':'), ensure_ascii=False, allow_nan=False, utf-8","leaf_hash":"sha256(0x00 || canonical_json(envelope_full))","seal_signature":"ed25519(seal minus {signature, sig_algorithm})"},"axiom_id":"axm_cf9fafbddc317965bcdb57308a396e5e8130f5f3fffd487367a1d6abe72bb9f4","bitcoin_anchor":{"bitcoin_attestations":["bitcoin_block_949451"],"calendar_attestations":["https://finney.calendar.eternitywall.com","https://btc.calendar.catallaxy.com","https://alice.btc.calendar.opentimestamps.org","https://bob.btc.calendar.opentimestamps.org"],"ots_url":"/registry/data/substrate/anchors/77fc9c28fae777b81da5b495b3115474df6592dfac590333213d3bdf8b94a9b3.ots","stamped_at":"2026-05-15T03:00:03Z","status":"bitcoin"},"envelope":{"anchors":[{"chain":"crovia.axiom_graph","height":0,"merkle_proof":"spider_vendor_press_v1","root_at_anchor":"spider_vendor_press_v1"}],"axiom_id":"axm_cf9fafbddc317965bcdb57308a396e5e8130f5f3fffd487367a1d6abe72bb9f4","axiom_type":"AX.OBS","body":{"axiom_subtype":"news.vendor_press.v1","category":"news","fingerprint":"bb8b31003b746fc1022f12c06e0188c6a90402161afeaba6aaae1400a47c55ef","published":"Tue, 05 May 2026 00:00:00 -0400","receipt_hash":"bb8b31003b746fc1022f12c06e0188c6a90402161afeaba6aaae1400a47c55ef","schema":"spider.news.vendor_press.v1","spider":"vendor_press","spider_record":{"axiom_subtype":"news.vendor_press.v1","category":"news","decision_hint":"POSITIVE","envelope_target":"AX.OBS","fingerprint":"bb8b31003b746fc1022f12c06e0188c6a90402161afeaba6aaae1400a47c55ef","observed_at":"2026-05-05T04:43:28.458286Z","parent_run_hash":"84f40dd8e6af9caa9f2a7fd632024f0e0192177c4532d9af5caf61b409324d7a","published":"Tue, 05 May 2026 00:00:00 -0400","runtime_version":"0.1.0","schema":"spider.news.vendor_press.v1","source_status":200,"source_url":"https://export.arxiv.org/rss/cs.AI","spider":"vendor_press","summary_excerpt":"arXiv:2605.00236v1 Announce Type: cross \nAbstract: Safety-aligned large language models rely on RLHF and instruction tuning to refuse harmful requests, yet the internal mechanisms implementing safety behavior remain poorly understood. We introduce the Attention Redistribution Attack (ARA), a white-box adversarial attack that identifies safety-critical attention heads and crafts nonsemantic adversarial tokens that redirect attention away from safety-relevant positions. Unlike prior jailbreak methods operating at the semantic or output-logit level, ARA targets the geometry of softmax attention on the probability simplex using Gumbel-softmax optimization over targeted heads. Across LLaMA-3-8B-Instruct, Mistral-7B-Instruct-v0.1, and Gemma-2-9B-it, ARA bypasses safety alignment with as few as 5 tokens and 500 optimization steps, achieving 36% ASR on Mistral-7B and 30% on LLaMA-3 against 200 HarmBench prompts, while Gemma-2 remains at 1%. Our principal mechanistic finding is a dissociation b","title":"Attention Is Where You Attack","url":"https://arxiv.org/abs/2605.00236","vendor":"arxiv_cs_ai"},"summary":"arXiv:2605.00236v1 Announce Type: cross \nAbstract: Safety-aligned large language models rely on RLHF and instruction tuning to refuse harmful requests, yet the internal mechanisms implementing safety behavior remain poorly understood. We introduce the Attention Redistribution Attack (ARA), a white-box adversarial attack that identifies safety-critical attention heads and crafts nonsemantic adversarial tokens that redirect attention away from safety-relevant positions. Unlike prior jailbreak methods operating at the semantic or output-logit level, ARA targets the geometry of softmax attention on the probability simplex using Gumbel-softmax optimization over targeted heads. Across LLaMA-3-8B-Instruct, Mistral-7B-Instruct-v0.1, and Gemma-2-9B-it, ARA bypasses safety alignment with as few as 5 tokens and 500 optimization steps, achieving 36% ASR on Mistral-7B and 30% on LLaMA-3 against 200 HarmBench prompts, while Gemma-2 remains at 1%. Our principal mechanistic finding is a dissociation b","title":"Attention Is Where You Attack","vendor":"arxiv_cs_ai"},"confidence":{"method":"deterministic"},"decision":"POSITIVE","issued_at":"2026-05-05T04:43:28Z","notes":"Spider vendor_press (news) news.vendor_press.v1","object":{"captured_by":"crovia.spider.vendor_press","primary_source_url":"https://arxiv.org/abs/2605.00236"},"predecessors":[],"schema":"crovia.axiom.v1","signature":"ed25519:a00e7dd55f7d25ecffc382007d0eab0e5d0153ac064243c5e800668fd9a3660e20b12307cb2cad6ed7424263c0457e77456f76495226ae8fa867b5ee8655a200","signer":"crovia.substrate","subject":{"observed_at":"2026-05-05T04:43:28Z","source_collector":"spider:vendor_press","target_id":"https://arxiv.org/abs/2605.00236"},"tsa":{"authority":"crovia.substrate.bootstrap","rfc3161_token":"{\"kind\":\"crovia.bootstrap.tsa\",\"source_jsonl\":\"/opt/crovia/spider/data/news/vendor_press_v1.jsonl\",\"source_seal_merkle_root\":\"spider_vendor_press_v1\",\"upgrade_path\":\"Sessione H \\u2014 OpenTimestamps weekly anchor\"}"},"zk_mode":"clear","zk_proof":null},"ledger":{"leaf_hash":"d493e359d4dda830e363399acf574da24d6727ac8aee6a62fd02fdf8220a60a3","leaf_index":114276,"ledger_path":"/opt/crovia/substrate/axiom_ledger.jsonl"},"merkle_proof":{"hash_alg":"sha256","leaf_prefix":"0x00","node_prefix":"0x01","odd_leaf_rule":"duplicate_last","path":[{"sibling":"cc7df7e8aa6ee60c063df6412b577443df9ec38a6f4ea72484c3cbb5ab51d17b","side":"right"},{"sibling":"1d714d0237b09a2d6a37f4094291c03b13dec95e199b0684ac5f294fb71ecc6d","side":"right"},{"sibling":"a15297c56668882861ed8fce644e636286f62a28ecb2ce72137bd3e122f48f91","side":"left"},{"sibling":"cf2da3c1e6fc4eacb93830e9169cd6ae0e3cf2d095bfcd1e8903246190e3168f","side":"right"},{"sibling":"6ba45046af77961de3939d5c39c0492561ede31871b5f9eb901aaa71a1ca9635","side":"right"},{"sibling":"1ad9ec90a9658a83c439eced513cb663f1576046e8ed981a3ed327a1e9106c96","side":"left"},{"sibling":"864ff8c61e24143ac1684572b26ed4ff162b3c2195104f55103383456befd161","side":"left"},{"sibling":"46dca532ded31bba3e38a2fc1bf625ddbe113eaf158d9999dfa7f7d5674bbb32","side":"right"},{"sibling":"17e21f83c86c2a86dddad911d22393ee81f41d8e0b79ef8c0c48668ea846f2ca","side":"right"},{"sibling":"8df6c69f776fbba620634f6baf125eeaa6fdf9426e461e649028010d9a703f42","side":"left"},{"sibling":"ea40d1bd0432ad4dd90f023c692aaab8c8f54e27ce652e82f2ec8e16cacfb632","side":"left"},{"sibling":"673cd6d27b696232f7f65e7a0733c7df6bc9fac3bcb569907ebb0de1c72c4bb2","side":"left"},{"sibling":"76d7c4385d71ae8104107105be66eddb792e01ed6e493db5a9b4eec5441abb76","side":"left"},{"sibling":"b03f005860bf95148ea89c703a32ca19ab7a2ca71b58bb628ff94a9edb8703b0","side":"left"},{"sibling":"21d1e30b556f553dc939fddc23e6367f0a7755ebe3dd489dc0001cef017beb7f","side":"right"},{"sibling":"f2817ab288b5324fe49770372c7a10f33f7cd11005f8d4c0a730316f5229dc98","side":"left"},{"sibling":"725fac972e772ca0dc598810ea1abc70df472f72d2d6ab8a0baee2b80e5d2f4c","side":"left"},{"sibling":"98fc57dfef8873b512edc8340f7181df57302bb96777625e072235c62d7c5895","side":"right"}]},"schema":"crovia.axiom_proof.v1","seal":{"first_collector_run_id":"","first_receipt_hash":"","jsonl_path":"/opt/crovia/substrate/axiom_ledger.jsonl","key_id":"430895f101d38164","last_collector_run_id":"","last_receipt_hash":"","leaf_count":134292,"merkle_root":"77fc9c28fae777b81da5b495b3115474df6592dfac590333213d3bdf8b94a9b3","public_key_hex":"cf742e26f75669dc673cb5c0786a1ae23ae8ca19c347317192ce40c28a7ff25c","run_id":"hourly_json_retrofit_20260515T023701Z","schema":"crovia.seal.v1","seal_family_version":"crovia-seal-family/1","seal_kind":"substrate_batch","sealed_at":"2026-05-15T02:37:25Z","sig_algorithm":"ed25519","signature":"68107a834b00b24f5d4501e5ec727445311f132a486567ecc4c72a4e6dff24c8c21f2de3105293353ba5fdbe370d032819af6aa70f694e2e39b6af6737507009","signer_version":"1.1.0"},"trust_root":{"key_id":"430895f101d38164","public_key_hex":"cf742e26f75669dc673cb5c0786a1ae23ae8ca19c347317192ce40c28a7ff25c","signature_algorithm":"ed25519","url":"/registry/canon/TRUST_ROOT.md"},"verifier":{"spec":"/registry/canon/AXIOM_RECEIPT_v1.md","url":"/v/axm_cf9fafbddc317965bcdb57308a396e5e8130f5f3fffd487367a1d6abe72bb9f4"}}