{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:3BT6KYVMF5JWIDWL3YPTU4PN5Q","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":"4b30d881f1cf3332097f0dc777d05f6157b3430ab07a46e1d77f206d6658f781","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-23T03:42:41Z","title_canon_sha256":"dc58c401d2e10dea15f01850d00bfb627f85b25b72306526e973e7d893b49b4a"},"schema_version":"1.0","source":{"id":"2604.21251","kind":"arxiv","version":5}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.21251","created_at":"2026-05-20T00:00:39Z"},{"alias_kind":"arxiv_version","alias_value":"2604.21251v5","created_at":"2026-05-20T00:00:39Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.21251","created_at":"2026-05-20T00:00:39Z"},{"alias_kind":"pith_short_12","alias_value":"3BT6KYVMF5JW","created_at":"2026-05-20T00:00:39Z"},{"alias_kind":"pith_short_16","alias_value":"3BT6KYVMF5JWIDWL","created_at":"2026-05-20T00:00:39Z"},{"alias_kind":"pith_short_8","alias_value":"3BT6KYVM","created_at":"2026-05-20T00:00:39Z"}],"graph_snapshots":[{"event_id":"sha256:80c45e7430bf3e1b2dcaadc5d84c2eab5713e210056abb15f46f8933717c0f31","target":"graph","created_at":"2026-05-20T00:00:39Z","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":"CAP achieves precise, controllable unlearning without updating model parameters, establishing a dynamic alignment mechanism that overcomes the transferability limitations of prior methods."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That reinforcement learning can train a prompt generator to collaborate with a fixed LLM such that target knowledge is suppressed while general capabilities remain selectively preserved and the effect is reversible upon prompt revocation."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"CAP is a reinforcement-learning-driven prompt optimization framework that suppresses target knowledge in LLMs while preserving general capabilities, enabling reversible unlearning without any parameter updates."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Reinforcement learning trains prompts that suppress specific knowledge in fixed LLMs while preserving general capabilities and allowing reversal by prompt removal."}],"snapshot_sha256":"8c389cb8520bffc1dc4f487684b4f64f2cae6155eb8d3a2e3f347e737b49f2f0"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"f999160140984ad363221ef326d77ff35166d22a5eb74c9407ce8cf6c209c2c9"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2604.21251/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large language models (LLMs) trained on unfiltered corpora inherently risk retaining sensitive information, necessitating selective knowledge unlearning for regulatory compliance and ethical safety. However, existing parameter-modifying methods face fundamental limitations: high computational costs, uncontrollable forgetting boundaries, and strict dependency on model weight access. These constraints render them impractical for closed-source models, yet current non-invasive alternatives remain unsystematic and reliant on empirical experience. To address these challenges, we propose the Controll","authors_text":"Guangchun Luo, Hongli Pu, Jie Ou, Jingwen Pu, Jinyu Guo, Meng Yang, Wenhong Tian, Wenyi Li, Xunlei Chen, Zhaokun Wang","cross_cats":["cs.AI"],"headline":"Reinforcement learning trains prompts that suppress specific knowledge in fixed LLMs while preserving general capabilities and allowing reversal by prompt removal.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-23T03:42:41Z","title":"CAP: Controllable Alignment Prompting for Unlearning in LLMs"},"references":{"count":9,"internal_anchors":5,"resolved_work":9,"sample":[{"cited_arxiv_id":"2403.04652","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Yi: Open Foundation Models by 01.AI","work_id":"8efee8a1-5e3c-4851-9c65-18e3d1d9e769","year":2024},{"cited_arxiv_id":"2412.19437","doi":"","is_internal_anchor":true,"ref_index":2,"title":"DeepSeek-V3 Technical Report","work_id":"57d2791d-2219-4c31-a077-afc04b12a75c","year":2024},{"cited_arxiv_id":"2406.12793","doi":"","is_internal_anchor":true,"ref_index":3,"title":"ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools","work_id":"de9ce5af-0d8d-4b94-9793-64968d9bc06d","year":2024},{"cited_arxiv_id":"2403.03218","doi":"","is_internal_anchor":true,"ref_index":4,"title":"The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning","work_id":"d05f8523-8089-4fdb-9c07-463952166528","year":2025},{"cited_arxiv_id":"2604.04942","doi":"","is_internal_anchor":true,"ref_index":5,"title":"TDA-RC: Task-Driven Alignment for Knowledge-Based Reasoning Chains in Large Language Models","work_id":"77051665-d34f-4db0-a66e-25fc43863478","year":2024}],"snapshot_sha256":"a7d4ae06d7fbf1248f65e8de2049f1d4782782b83cf652dae6404d9b9d53c360"},"source":{"id":"2604.21251","kind":"arxiv","version":5},"verdict":{"created_at":"2026-05-19T17:09:46.710788Z","id":"122d6b06-2259-48eb-8386-3ea584f94e1a","model_set":{"reader":"grok-4.3"},"one_line_summary":"CAP is a reinforcement-learning-driven prompt optimization framework that suppresses target knowledge in LLMs while preserving general capabilities, enabling reversible unlearning without any parameter updates.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Reinforcement learning trains prompts that suppress specific knowledge in fixed LLMs while preserving general capabilities and allowing reversal by prompt removal.","strongest_claim":"CAP achieves precise, controllable unlearning without updating model parameters, establishing a dynamic alignment mechanism that overcomes the transferability limitations of prior methods.","weakest_assumption":"That reinforcement learning can train a prompt generator to collaborate with a fixed LLM such that target knowledge is suppressed while general capabilities remain selectively preserved and the effect is reversible upon prompt revocation."}},"verdict_id":"122d6b06-2259-48eb-8386-3ea584f94e1a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c3506bde2ecf062438fb070819b84f10aa51006ec4327fe2490f8db4d7013b9c","target":"record","created_at":"2026-05-20T00:00:39Z","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":"4b30d881f1cf3332097f0dc777d05f6157b3430ab07a46e1d77f206d6658f781","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-23T03:42:41Z","title_canon_sha256":"dc58c401d2e10dea15f01850d00bfb627f85b25b72306526e973e7d893b49b4a"},"schema_version":"1.0","source":{"id":"2604.21251","kind":"arxiv","version":5}},"canonical_sha256":"d867e562ac2f53640ecbde1f3a71edec3e23cfb031c3fe0d3256efdc1d6db991","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"d867e562ac2f53640ecbde1f3a71edec3e23cfb031c3fe0d3256efdc1d6db991","first_computed_at":"2026-05-20T00:00:39.303678Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:00:39.303678Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"PjaJ8OPyeOAy8i0omKL+4DWos2whrjFLaDORD/MBskztKJqgjECdrLDrrf/TAx6d6SJxEd2dou2Mv5SwX3eDBw==","signature_status":"signed_v1","signed_at":"2026-05-20T00:00:39.304387Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.21251","source_kind":"arxiv","source_version":5}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c3506bde2ecf062438fb070819b84f10aa51006ec4327fe2490f8db4d7013b9c","sha256:80c45e7430bf3e1b2dcaadc5d84c2eab5713e210056abb15f46f8933717c0f31"],"state_sha256":"228d86c4b9d6accf8a6a33891965b9c847555c4ed45dbf8d255e079779d023f6"}