{"_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_166cc6ccf2841ebc84bdf4990bd7dadbdd04d2b7647ed426742b16ec746326ac","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_166cc6ccf2841ebc84bdf4990bd7dadbdd04d2b7647ed426742b16ec746326ac","axiom_type":"AX.OBS","body":{"axiom_subtype":"news.vendor_press.v1","category":"news","fingerprint":"7fa8214e275b06cb9f47135d8905c8d7a62990491b933cbbc17d30dadcdb876a","published":"Fri, 08 May 2026 00:00:00 -0400","receipt_hash":"7fa8214e275b06cb9f47135d8905c8d7a62990491b933cbbc17d30dadcdb876a","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":"7fa8214e275b06cb9f47135d8905c8d7a62990491b933cbbc17d30dadcdb876a","observed_at":"2026-05-08T04:43:40.537619Z","parent_run_hash":"9837a17a0d4866b3bef2929e933ca29d96f4bd9656766df36f8260d720835b95","published":"Fri, 08 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:2605.05415v1 Announce Type: cross \nAbstract: Large language models (LLMs) remain vulnerable to adversarial prompting despite advances in alignment and safety, often exhibiting harmful behaviors under novel attack strategies. While adversarial training can improve robustness, existing approaches are computationally expensive and difficult to scale. Recent continuous adversarial training methods, such as Continuous adversarial training (CAT) and Continuous Adversarial Preference Optimization (CAPO), address this challenge by leveraging gradient-based perturbations in the embedding space, enabling more efficient and expressive attacks. Building on this paradigm, we propose WARDEN, a distributionally robust adversarial training framework for LLMs that dynamically reweights adversarial examples through an f -divergence ambiguity set around the empirical training distribution. Our method optimizes the worst-case adversarial loss within a divergence ball around the empirical data distri","title":"Information Theoretic Adversarial Training of Large Language Models","url":"https://arxiv.org/abs/2605.05415","vendor":"arxiv_cs_ai"},"summary":"arXiv:2605.05415v1 Announce Type: cross \nAbstract: Large language models (LLMs) remain vulnerable to adversarial prompting despite advances in alignment and safety, often exhibiting harmful behaviors under novel attack strategies. While adversarial training can improve robustness, existing approaches are computationally expensive and difficult to scale. Recent continuous adversarial training methods, such as Continuous adversarial training (CAT) and Continuous Adversarial Preference Optimization (CAPO), address this challenge by leveraging gradient-based perturbations in the embedding space, enabling more efficient and expressive attacks. Building on this paradigm, we propose WARDEN, a distributionally robust adversarial training framework for LLMs that dynamically reweights adversarial examples through an f -divergence ambiguity set around the empirical training distribution. Our method optimizes the worst-case adversarial loss within a divergence ball around the empirical data distri","title":"Information Theoretic Adversarial Training of Large Language Models","vendor":"arxiv_cs_ai"},"confidence":{"method":"deterministic"},"decision":"POSITIVE","issued_at":"2026-05-08T04:43:40Z","notes":"Spider vendor_press (news) news.vendor_press.v1","object":{"captured_by":"crovia.spider.vendor_press","primary_source_url":"https://arxiv.org/abs/2605.05415"},"predecessors":[],"schema":"crovia.axiom.v1","signature":"ed25519:6680adc8153aae67348812418925bb368e559fb3934bcd9fbcd0dbdb8715c6409cc5daf3511c6edeca7cf28c763ec929c7c50cfa454381cf324b9755e052500f","signer":"crovia.substrate","subject":{"observed_at":"2026-05-08T04:43:40Z","source_collector":"spider:vendor_press","target_id":"https://arxiv.org/abs/2605.05415"},"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":"bb8809b1225bf62fbc632ef8576301354763cfe834f9e1533255b74133adf4e7","leaf_index":120189,"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":"df8128a9b8ecdeaf6f162eaba3cb99b0242dd71fa3ce2cb0b1f20072cdd96f57","side":"left"},{"sibling":"0d8b53c61e078bb55c6c642ec167202abcf70d1239b3ccc2abcf208cf845c595","side":"right"},{"sibling":"7c7d7f33fde4d38c1d72b8cecff077a75c70d6fbc32e59e57285f4498daadf15","side":"left"},{"sibling":"428eb298d3bd2a3552abf1946010e0392b970809f1985739aeb1bcc502aa183c","side":"left"},{"sibling":"590df1d67ad58cf979531ca8f1898a26a853d9042c1d9828c89dcebfd1381d1f","side":"left"},{"sibling":"55c2fe9ae9c677eb542780acf595acf6df2e8cf81924322ae0460b003293be7a","side":"left"},{"sibling":"fbc97c8ffc4e0529c02cda4ab186b044b09b474182fb9fbd8e687b59c97e08b2","side":"left"},{"sibling":"b1a469ecc2c62da181fb78fb052d1f4d059aafbe92dbb6d04682090381d419fd","side":"right"},{"sibling":"34e93c6f591cc4fb93a4c29771210eb73573b454d2c5329963f437f1309fe46f","side":"left"},{"sibling":"b9834433f5bd1deeaab4ced2b3bcc0d91d19763a5d0981fb299d124b63459eb3","side":"right"},{"sibling":"143f33d3924b3840fd6dd8ba12566fc35bf86189e663ef0ad4884d676f295e3a","side":"left"},{"sibling":"b1ed99341c327c7c9ab2489f40af2547ab3b3b4b6a74fb684210164fb891a413","side":"right"},{"sibling":"6ee3be9bdfc9bee55d32f7dbb0075f02fe87d20887d563d3e300caf36b1b88c7","side":"left"},{"sibling":"8ccd9937a2c0d5c04044d07d1557791b7d07bb31eac41a39a675608d44b38f23","side":"right"},{"sibling":"3a5e69cf0803f4c91f3895ed7c9a95748fef240bec4422e167c05300f79f06c0","side":"left"},{"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_166cc6ccf2841ebc84bdf4990bd7dadbdd04d2b7647ed426742b16ec746326ac"}}