{"_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_3f3302936dee7eca83f7092ead5381cee1087e6678fc28f1fab86abaf4fe091b","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_3f3302936dee7eca83f7092ead5381cee1087e6678fc28f1fab86abaf4fe091b","axiom_type":"AX.OBS","body":{"axiom_subtype":"news.vendor_press.v1","category":"news","fingerprint":"e4dbe69109e452c051d3bb9e7f97a0535178e61137fa944d58b89296e14c559c","published":"Wed, 06 May 2026 00:00:00 -0400","receipt_hash":"e4dbe69109e452c051d3bb9e7f97a0535178e61137fa944d58b89296e14c559c","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":"e4dbe69109e452c051d3bb9e7f97a0535178e61137fa944d58b89296e14c559c","observed_at":"2026-05-06T04:43:18.518819Z","parent_run_hash":"184209fc0f2ebdc4b721e1a74f74ab17637dab3344682b6e6bd1f3e5e8f2cd00","published":"Wed, 06 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:2605.02269v1 Announce Type: new \nAbstract: Specification gaming is a critical failure mode of LLM agents. Despite this, there has been little systematic research into when it arises and what drives it. To address this, we build and open source a diverse suite of tasks where models can score highly by taking unintended actions. We find that all tested models exploit their specifications at non-negligible rates in most of our eight settings, including five non-coding settings. We see the highest rates of specification gaming in Grok 4 and the lowest rates in Claude models. We use our evaluation suite to study what drives specification gaming, and find that: 1. RL reasoning training substantially increases the rate at which models exploit their specifications, 2. Increasing RL reasoning budget has a weakly positive effect on exploit rate, and 3. Test-time mitigations reduce but do not eliminate the rate of specification gaming. Our results suggest that specification gaming is a fund","title":"Towards Understanding Specification Gaming in Reasoning Models","url":"https://arxiv.org/abs/2605.02269","vendor":"arxiv_cs_ai"},"summary":"arXiv:2605.02269v1 Announce Type: new \nAbstract: Specification gaming is a critical failure mode of LLM agents. Despite this, there has been little systematic research into when it arises and what drives it. To address this, we build and open source a diverse suite of tasks where models can score highly by taking unintended actions. We find that all tested models exploit their specifications at non-negligible rates in most of our eight settings, including five non-coding settings. We see the highest rates of specification gaming in Grok 4 and the lowest rates in Claude models. We use our evaluation suite to study what drives specification gaming, and find that: 1. RL reasoning training substantially increases the rate at which models exploit their specifications, 2. Increasing RL reasoning budget has a weakly positive effect on exploit rate, and 3. Test-time mitigations reduce but do not eliminate the rate of specification gaming. Our results suggest that specification gaming is a fund","title":"Towards Understanding Specification Gaming in Reasoning Models","vendor":"arxiv_cs_ai"},"confidence":{"method":"deterministic"},"decision":"POSITIVE","issued_at":"2026-05-06T04:43:18Z","notes":"Spider vendor_press (news) news.vendor_press.v1","object":{"captured_by":"crovia.spider.vendor_press","primary_source_url":"https://arxiv.org/abs/2605.02269"},"predecessors":[],"schema":"crovia.axiom.v1","signature":"ed25519:94f75dc390e73130301a4fd632bc0a86f4dafac50877ba2201a722e21f655c397c7c1e8884384ab560e01acc13edf405f65cc46c697181e8f8d94ab7168db106","signer":"crovia.substrate","subject":{"observed_at":"2026-05-06T04:43:18Z","source_collector":"spider:vendor_press","target_id":"https://arxiv.org/abs/2605.02269"},"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":"31f2de50ce591a278ac59fe5a945c9d81ddb89a8a24a1f435b3ce0e1da39f93c","leaf_index":116093,"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":"571db8dcdb83d7913207f72ae2a532f496cec405929864493b28be281ebb8d17","side":"left"},{"sibling":"5305ce80369f00d5e16a0aa972fd8bef9a44d006e9bf16ef3de44b50a3a9a48d","side":"right"},{"sibling":"e63f9527ab5b772af0d39bff189f0c6ddb2336c8abce87ee50e8f44c9eb5ec7e","side":"left"},{"sibling":"8293aba6c2475b373bedc1c69d9f9d1012ffebab80598d682b6c1ee7a53247d2","side":"left"},{"sibling":"e0fe8f9b44a3d6cbadab508f86e5c9d55ad530e3c089e5dcc25ffed0c325d61e","side":"left"},{"sibling":"3971b54bdd4f12d9e802c8187ad5d1ca18e58ad052ea54f133d256ed9e671f72","side":"left"},{"sibling":"f2fc4c75d0d98883e6651d6b8de4d8a9e32629c30756d80e57438acc15bc7799","side":"left"},{"sibling":"c8f2d2d59982877f539ec7300b3050bbd7090f678e1ec0fe9c92da90a24d82fa","side":"right"},{"sibling":"aece872b65a68d3775a2636761a78e8ecbdc28b94538a659e6069c57bef430e0","side":"left"},{"sibling":"fb3602a0ad8c9fcab7b18e13690c40ca4a2f5402557d976d681a369555c04ae2","side":"right"},{"sibling":"ddc060ac400459417799f688c75d5b271636afa40d89ec7fc98652473a8061f6","side":"left"},{"sibling":"282afa51266e47629e34d808a360bbb276348bcc5a24bdcc93e83a76e580293e","side":"right"},{"sibling":"05f89b32c00462e60adf95c1fe4579cdc2791b36e8b17573d8f3b5fd5da95a0b","side":"right"},{"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_3f3302936dee7eca83f7092ead5381cee1087e6678fc28f1fab86abaf4fe091b"}}