{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreidamj2nxd5go2wubbyaqhvmme4dedx2piiqjiwz6n5nkz2yoxpg3a",
"uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mjlazebbkpz2"
},
"path": "/t/seeking-arxiv-cs-ai-endorsement-refusal-first-memory-for-ai-agents/175294#post_1",
"publishedAt": "2026-04-15T23:39:23.000Z",
"site": "https://discuss.huggingface.co",
"tags": [
"GitHub - Preston2012/demi: Trust-first memory system for AI agents. MCP + REST. SQLite. $6/mo. BEAM #2 worldwide. · GitHub",
"Log in to arXiv | arXiv e-print repository"
],
"textContent": "I am an independent researcher seeking endorsement for my first arXiv submission in cs.AI. The paper evaluates an adaptive memory system across three benchmarks (LOCOMO, BEAM, LongMemEval) with two main findings: (1) answer model sensitivity varies by benchmark, with three models producing identical LOCOMO scores on the same retrieval pipeline, and (2) a refusal-first write pipeline blocks 19/19 adversarial attack vectors with 0.00% false positives. Full paper and MIT-licensed code: GitHub - Preston2012/demi: Trust-first memory system for AI agents. MCP + REST. SQLite. $6/mo. BEAM #2 worldwide. · GitHub. My arXiv endorsement code is WTOYS6 (enter at Log in to arXiv | arXiv e-print repository). If you are an arXiv endorser in cs.AI and would be willing to review and endorse, please reply or DM. Thank you!",
"title": "Seeking arXiv cs.AI endorsement: Refusal-First Memory for AI Agents"
}