{"assertions":[{"assertion_index":0,"assertion_text":"Back then it pitched its 2.6-billion-parameter model as competitive with models four times its size on tool use and multi-step agent tasks, and its 3-billion vision model as level with two rival small models, Gemma-4-E2B and Qwen3.5-2B, on tool-use tests Hugging Face Blog Hugging Face Blog.","citation_check":null,"claim_ids":[42145,42146,42147,42156,42157,45792,45793,45794,45795,45796,45797,45798],"link_only_urls":[]},{"assertion_index":1,"assertion_text":"Liquid concedes that larger models still hold the edge on knowledge-heavy tests such as GPQA Diamond, HLE, and AA-Omniscience, which measure what a model knows rather than what it can do Hugging Face Blog.","citation_check":null,"claim_ids":[77999,78000,78001,78002,78003,78004],"link_only_urls":[]},{"assertion_index":2,"assertion_text":"Liquid reports 184 tokens (the word fragments a model reads and produces) per second on an Apple M5 Max device and up to 21,000 per second on a single Nvidia B300 server chip, and in August its 2.6B model managed 220 per second on the same class of M5 Max chip while the 3B vision model hit 228 Hugging Face Blog Hugging Face Blog Hugging Face Blog.","citation_check":null,"claim_ids":[42145,42146,42147,42156,42157,45792,45793,45794,45795,45796,45797,45798,77999,78000,78001,78002,78003,78004],"link_only_urls":[]},{"assertion_index":3,"assertion_text":"Ollama's latest release candidate (a test build ahead of a stable version) ports Apple-specific kernels, the low-level code that runs the math on a chip, to CUDA, Nvidia's chip software, and llama.cpp, another free local-model runner, just fixed a crash where machines reporting zero free memory produced invalid layer splits and aborted loading Ollama Releases llama.cpp Releases.","citation_check":null,"claim_ids":[78116,78117,78445],"link_only_urls":[]},{"assertion_index":4,"assertion_text":"The firm pointed 25 package managers (the tools that pull code libraries into a project) at a compromised repository, and most of them would not catch a swapped file by default, because fewer than half ship with a lockfile (the record of the exact versions a project expects) that checks for tampering, and only two of the 25 checked every single download they pulled Trail of Bits.","citation_check":null,"claim_ids":[78170,78171,78172],"link_only_urls":[]},{"assertion_index":5,"assertion_text":"A self-spreading worm that began with one hijacked maintainer account reached more than 400 packages on npm, the main public store of JavaScript code libraries Microsoft Security Blog Wiz Research.","citation_check":null,"claim_ids":[42082,42083,42094,42582,42594,42615],"link_only_urls":[]},{"assertion_index":6,"assertion_text":"On access, Google is widening experimental use of its Geospatial Reasoning agent through Ask Google Earth and an Agent-to-Agent service, a protocol that lets your own agent hand a question to Google's, and the agent lets Gemini query Google's weather, flood, population, and satellite models alongside its mapping and public datasets Google Research Blog.","citation_check":null,"claim_ids":[78345,78346,78347],"link_only_urls":[]},{"assertion_index":7,"assertion_text":"On cost, Ai2's HiRO-ACEv2 simulates rare rain, wind, and pressure extremes at 3 km resolution for about 1/300th the cost of the X-SHiELD atmospheric model Allen Institute for AI.","citation_check":null,"claim_ids":[78134,78135,78136],"link_only_urls":[]},{"assertion_index":8,"assertion_text":"Claude already handles that narrow slice with a specific behavior, a banner that sends voters to TurboVote, and that is the useful part for anyone outside Anthropic: the company has a stated, election-specific response ready for the 0.51 percent Anthropic News Anthropic News Princeton CITP - Freedom to Tinker.","citation_check":null,"claim_ids":[75170,75174,75286,78038,78039,78040],"link_only_urls":[]},{"assertion_index":9,"assertion_text":"- Liquid AI's five-of-eight lead-or-tie on tool-use tests is self-reported; independent reruns of LiveClaw and BFCLv4 on the 24B-A2B model will show whether the on-device number travels. Hugging Face Blog","citation_check":null,"claim_ids":[77999,78000,78001,78002,78003,78004],"link_only_urls":[]},{"assertion_index":10,"assertion_text":"- Ollama's CUDA port of its Apple-chip kernels is still a release candidate; a stable 0.40.5 would bring the ported Apple kernels to Nvidia machines. Ollama Releases","citation_check":null,"claim_ids":[78445],"link_only_urls":[]},{"assertion_index":11,"assertion_text":"- Trail of Bits found only two package managers that verify every download; the next move is which ecosystems turn integrity checks on by default. Trail of Bits","citation_check":null,"claim_ids":[78170,78171,78172],"link_only_urls":[]},{"assertion_index":12,"assertion_text":"- Google calls Geospatial Reasoning access experimental; the Augur call above tracks whether that becomes general availability by March. Google Research Blog","citation_check":null,"claim_ids":[78345,78346,78347],"link_only_urls":[]},{"assertion_index":13,"assertion_text":"- Anthropic's 0.51 percent election share is a mid-October baseline; a follow-up count closer to the vote would show whether queries climb. Anthropic News","citation_check":null,"claim_ids":[78038,78039,78040],"link_only_urls":[]},{"assertion_index":14,"assertion_text":"Every benchmark and speed figure for LFM2.5-24B-A2B comes from Liquid AI's own post and is unaudited in the evidence, and the LiveClaw comparison is against a single named rival Hugging Face Blog.","citation_check":null,"claim_ids":[77999,78000,78001,78002,78003,78004],"link_only_urls":[]}],"brief_date":"2026-10-15","bundle_id":"evidence-bundle-v2-6817df1bd29bbba7e6a3c7995c46cef3036486c5d28c96c13a7a5b6e35ded2dc","claims":[{"assertion_indexes":[0,2],"citation_authority_claim_id":42145,"citation_checks":[],"claim_id":42145,"claim_text":"Liquid AI claims LFM2.5-2.6B is competitive with models four times its size on tool use, instruction following, and multi-step agentic tasks.","correction_role":null,"corroborates_eligible":true,"label":"opinion","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:19:49.595841+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-2-6b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":42146,"citation_checks":[],"claim_id":42146,"claim_text":"LFM2.5-2.6B achieves an inference speed of 220 tokens per second on an Apple M5 Max.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:19:49.595841+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-2-6b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":42147,"citation_checks":[],"claim_id":42147,"claim_text":"LFM2.5-2.6B achieves an inference speed of 113 tokens per second on an AMD Ryzen CPU.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:19:49.595841+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-2-6b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":42156,"citation_checks":[],"claim_id":42156,"claim_text":"Liquid AI claims LFM2.5-2.6B is the fastest model tested on CPU inference.","correction_role":null,"corroborates_eligible":true,"label":"opinion","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:19:49.595841+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-2-6b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":42157,"citation_checks":[],"claim_id":42157,"claim_text":"Liquid AI claims LFM2.5-2.6B reaches almost 15,000 output tokens per second on an H100 GPU.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:19:49.595841+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-2-6b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":45792,"citation_checks":[],"claim_id":45792,"claim_text":"Liquid AI reports that LFM2.5-VL-3B achieves an average score of 69.4 on its combined vision benchmark suite.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-13T06:25:57.869622+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-vl-3b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":45793,"citation_checks":[],"claim_id":45793,"claim_text":"Liquid AI asserts that LFM2.5-VL-3B performs on par with Gemma-4-E2B and Qwen3.5-2B on tool use benchmarks.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-13T06:25:57.869622+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-vl-3b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":45794,"citation_checks":[],"claim_id":45794,"claim_text":"Liquid AI states that LFM2.5-VL-3B decodes at 228 tokens per second on an M5 Max chip.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-13T06:25:57.869622+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-vl-3b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":45795,"citation_checks":[],"claim_id":45795,"claim_text":"Liquid AI claims that LFM2.5-VL-3B decodes at 116 tokens per second on a Ryzen AI Max+ 395.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-13T06:25:57.869622+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-vl-3b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":45796,"citation_checks":[],"claim_id":45796,"claim_text":"Liquid AI states that LFM2.5-VL-3B reaches 20 tokens per second on a Galaxy S26 Ultra, enabling fully on-device inference.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-13T06:25:57.869622+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-vl-3b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":45797,"citation_checks":[],"claim_id":45797,"claim_text":"Liquid AI reports that LFM2.5-VL-3B achieves an output throughput of approximately 11,000 tokens per second at high concurrency on an H100 GPU.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-13T06:25:57.869622+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-vl-3b"},{"assertion_indexes":[0,2],"citation_authority_claim_id":45798,"citation_checks":[],"claim_id":45798,"claim_text":"Liquid AI claims that the 11,000 tokens per second throughput is roughly twice that of larger 4B-class models and ahead of smaller 2B-class models.","correction_role":null,"corroborates_eligible":true,"label":"opinion","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-13T06:25:57.869622+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm2-5-vl-3b"},{"assertion_indexes":[1,2,9,14],"citation_authority_claim_id":77999,"citation_checks":[],"claim_id":77999,"claim_text":"Liquid AI's LFM2.5-24B-A2B model scores 48.09 on the LiveClaw benchmark, outperforming Qwen3.6-35B-A3B's score of 38.19.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:45:23.385614+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b"},{"assertion_indexes":[1,2,9,14],"citation_authority_claim_id":78000,"citation_checks":[],"claim_id":78000,"claim_text":"Liquid AI claims LFM2.5-24B-A2B achieves an inference speed of 184 tokens per second on an Apple M5 Max device.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:45:23.385614+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b"},{"assertion_indexes":[1,2,9,14],"citation_authority_claim_id":78001,"citation_checks":[],"claim_id":78001,"claim_text":"Liquid AI states that LFM2.5-24B-A2B reaches up to 21,000 output tokens per second on a single NVIDIA B300 GPU.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:45:23.385614+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b"},{"assertion_indexes":[1,2,9,14],"citation_authority_claim_id":78002,"citation_checks":[],"claim_id":78002,"claim_text":"Liquid AI reports that LFM2.5-24B-A2B leads or matches competitors on five of eight tool-use benchmarks, specifically AutomationBench, BFCLv4, ToolSandbox, LiveClaw, and \u03c4\u00b3-Bench Banking.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:45:23.385614+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b"},{"assertion_indexes":[1,2,9,14],"citation_authority_claim_id":78003,"citation_checks":[],"claim_id":78003,"claim_text":"Liquid AI asserts that larger models maintain an advantage over LFM2.5-24B-A2B on knowledge-heavy benchmarks such as GPQA Diamond, HLE, and AA-Omniscience.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:45:23.385614+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b"},{"assertion_indexes":[1,2,9,14],"citation_authority_claim_id":78004,"citation_checks":[],"claim_id":78004,"claim_text":"Liquid AI characterizes LFM2.5-24B-A2B as a strong foundation for agents requiring reasoning, tool calling, and reliable on-device execution.","correction_role":null,"corroborates_eligible":true,"label":"opinion","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:45:23.385614+00:00","source_name":"Hugging Face Blog","source_url":"https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b"},{"assertion_indexes":[3],"citation_authority_claim_id":78116,"citation_checks":[],"claim_id":78116,"claim_text":"Prior to the b11626 release, when all devices reported zero free memory, split points were normalized by a zero sum, causing NaN splits that sent layers to invalid device indices and aborted loading with a vector::_M_range_check error.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:50:08.961582+00:00","source_name":"llama.cpp Releases","source_url":"https://github.com/ggml-org/llama.cpp/releases/tag/b11626"},{"assertion_indexes":[3],"citation_authority_claim_id":78117,"citation_checks":[],"claim_id":78117,"claim_text":"The b11626 release of llama.cpp implements a fix to split layers evenly when no device reports free memory, addressing the previous zero-sum normalization issue.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:50:08.961582+00:00","source_name":"llama.cpp Releases","source_url":"https://github.com/ggml-org/llama.cpp/releases/tag/b11626"},{"assertion_indexes":[3,10],"citation_authority_claim_id":78445,"citation_checks":[],"claim_id":78445,"claim_text":"Ollama version v0.40.5-rc0 includes a port of MLX Metal custom kernels to CUDA by @dhiltgen.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T07:02:09.495059+00:00","source_name":"Ollama Releases","source_url":"https://github.com/ollama/ollama/releases/tag/v0.40.5-rc0"},{"assertion_indexes":[4,11],"citation_authority_claim_id":78170,"citation_checks":[],"claim_id":78170,"claim_text":"The author tested 25 popular package managers for their ability to detect compromised repositories injecting malicious code.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:52:03.352590+00:00","source_name":"Trail of Bits","source_url":"https://blog.trailofbits.com/2026/10/14/package-managers-have-an-integrity-problem/"},{"assertion_indexes":[4,11],"citation_authority_claim_id":78171,"citation_checks":[],"claim_id":78171,"claim_text":"Fewer than half of the 25 tested package managers used lockfiles that checked for tampering by default.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:52:03.352590+00:00","source_name":"Trail of Bits","source_url":"https://blog.trailofbits.com/2026/10/14/package-managers-have-an-integrity-problem/"},{"assertion_indexes":[4,11],"citation_authority_claim_id":78172,"citation_checks":[],"claim_id":78172,"claim_text":"Only two of the 25 tested package managers verified the integrity of every package they downloaded.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:52:03.352590+00:00","source_name":"Trail of Bits","source_url":"https://blog.trailofbits.com/2026/10/14/package-managers-have-an-integrity-problem/"},{"assertion_indexes":[5],"citation_authority_claim_id":42082,"citation_checks":[],"claim_id":42082,"claim_text":"The attack resulted from the compromise of a GitHub maintainer account, which led to the publication of malicious package versions.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:18:51.975110+00:00","source_name":"Wiz Research","source_url":"https://www.wiz.io/blog/keyv-and-cacheable-npm-supply-chain-attack"},{"assertion_indexes":[5],"citation_authority_claim_id":42083,"citation_checks":[],"claim_id":42083,"claim_text":"The malicious worm propagated to over 400 distinct npm packages.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:18:51.975110+00:00","source_name":"Wiz Research","source_url":"https://www.wiz.io/blog/keyv-and-cacheable-npm-supply-chain-attack"},{"assertion_indexes":[5],"citation_authority_claim_id":42094,"citation_checks":[],"claim_id":42094,"claim_text":"Wiz advises treating affected systems as potentially compromised and rebuilding them if malicious packages were installed.","correction_role":null,"corroborates_eligible":true,"label":"opinion","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:18:51.975110+00:00","source_name":"Wiz Research","source_url":"https://www.wiz.io/blog/keyv-and-cacheable-npm-supply-chain-attack"},{"assertion_indexes":[5],"citation_authority_claim_id":42582,"citation_checks":[],"claim_id":42582,"claim_text":"Microsoft Threat Intelligence identified a large-scale npm supply chain attack affecting more than 400 packages.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-06T06:22:01.138762+00:00","source_name":"Microsoft Security Blog","source_url":"https://www.microsoft.com/en-us/security/blog/2026/08/04/chaindrop-supply-chain-compromise-anatomy-self-propagating-worm/"},{"assertion_indexes":[5],"citation_authority_claim_id":42594,"citation_checks":[],"claim_id":42594,"claim_text":"Many malicious versions had no corresponding source-code commit, pull request, tag, or legitimate release.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-06T06:22:01.138762+00:00","source_name":"Microsoft Security Blog","source_url":"https://www.microsoft.com/en-us/security/blog/2026/08/04/chaindrop-supply-chain-compromise-anatomy-self-propagating-worm/"},{"assertion_indexes":[5],"citation_authority_claim_id":42615,"citation_checks":[],"claim_id":42615,"claim_text":"This mechanism can rapidly transform one compromised npm identity into many malicious package releases.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-06T06:22:01.138762+00:00","source_name":"Microsoft Security Blog","source_url":"https://www.microsoft.com/en-us/security/blog/2026/08/04/chaindrop-supply-chain-compromise-anatomy-self-propagating-worm/"},{"assertion_indexes":[6,12],"citation_authority_claim_id":78345,"citation_checks":[],"claim_id":78345,"claim_text":"Google Research is expanding experimental access to its Geospatial Reasoning agent via Ask Google Earth and an Agent-to-Agent (A2A) service.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:59:21.019912+00:00","source_name":"Google Research Blog","source_url":"https://research.google/blog/putting-earth-ai-to-work-geospatial-agents-for-our-planets-biggest-challenges/"},{"assertion_indexes":[6,12],"citation_authority_claim_id":78346,"citation_checks":[],"claim_id":78346,"claim_text":"The Geospatial Reasoning agent connects Gemini to Earth AI foundation models including WeatherNext, Flood Forecasting, the Population Dynamics Foundation Model (PDFM), AlphaEarth Foundations, and Remote Sensing Foundations.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:59:21.019912+00:00","source_name":"Google Research Blog","source_url":"https://research.google/blog/putting-earth-ai-to-work-geospatial-agents-for-our-planets-biggest-challenges/"},{"assertion_indexes":[6,12],"citation_authority_claim_id":78347,"citation_checks":[],"claim_id":78347,"claim_text":"The Geospatial Reasoning agent also integrates information from Google Maps Platform, Google Earth Engine, Data Commons, and Google Dataset Search.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:59:21.019912+00:00","source_name":"Google Research Blog","source_url":"https://research.google/blog/putting-earth-ai-to-work-geospatial-agents-for-our-planets-biggest-challenges/"},{"assertion_indexes":[7],"citation_authority_claim_id":78134,"citation_checks":[],"claim_id":78134,"claim_text":"Allen AI introduced HiRO-ACEv2, an update to its High Resolution Output Ai2 Climate Emulator system.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:51:14.688070+00:00","source_name":"Allen Institute for AI","source_url":"https://allenai.org/blog/hiro-acev2"},{"assertion_indexes":[7],"citation_authority_claim_id":78135,"citation_checks":[],"claim_id":78135,"claim_text":"HiRO-ACEv2 can reproduce extremely rare rain, wind, and pressure events with high accuracy.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:51:14.688070+00:00","source_name":"Allen Institute for AI","source_url":"https://allenai.org/blog/hiro-acev2"},{"assertion_indexes":[7],"citation_authority_claim_id":78136,"citation_checks":[],"claim_id":78136,"claim_text":"The HiRO-ACEv2 system generates global 3 km climate data at approximately 1/300th the cost of the X-SHiELD atmospheric model.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:51:14.688070+00:00","source_name":"Allen Institute for AI","source_url":"https://allenai.org/blog/hiro-acev2"},{"assertion_indexes":[8],"citation_authority_claim_id":75170,"citation_checks":[],"claim_id":75170,"claim_text":"Anthropic updated its Usage Policy to clarify rules regarding Claude's long-term, independent work capabilities and new misuse patterns in influence operations, weapons development, and surveillance.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-09T06:47:27.853171+00:00","source_name":"Anthropic News","source_url":"https://www.anthropic.com/news/2026-usage-policy-update"},{"assertion_indexes":[8],"citation_authority_claim_id":75174,"citation_checks":[],"claim_id":75174,"claim_text":"Anthropic updated its surveillance policy to explicitly prohibit tracking people without consent and using Claude to decide or recommend law enforcement actions like arrests.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-09T06:47:27.853171+00:00","source_name":"Anthropic News","source_url":"https://www.anthropic.com/news/2026-usage-policy-update"},{"assertion_indexes":[8],"citation_authority_claim_id":75286,"citation_checks":[],"claim_id":75286,"claim_text":"Anthropic\u2019s Claude provides an election banner directing users to TurboVote and strives for political even-handedness by avoiding unsolicited political opinions.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-09T06:52:18.156784+00:00","source_name":"Princeton CITP - Freedom to Tinker","source_url":"https://blog.citp.princeton.edu/2026/10/08/ai-election-policy/"},{"assertion_indexes":[8,13],"citation_authority_claim_id":78038,"citation_checks":[],"claim_id":78038,"claim_text":"Anthropic analyzed 250,000 Claude conversations over a one-week period using Anthropic Insights to understand how people use the model for elections and politics. (fact, 2026-10-14T00:00:00+00:00, null, null)","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:47:55.707841+00:00","source_name":"Anthropic News","source_url":"https://www.anthropic.com/news/2026-us-midterms"},{"assertion_indexes":[8,13],"citation_authority_claim_id":78039,"citation_checks":[],"claim_id":78039,"claim_text":"Approximately 5.75% of all Claude conversations in the United States touch broadly on politics, public policy, or social issues. (fact, 2026-10-14T00:00:00+00:00, null, null)","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:47:55.707841+00:00","source_name":"Anthropic News","source_url":"https://www.anthropic.com/news/2026-us-midterms"},{"assertion_indexes":[8,13],"citation_authority_claim_id":78040,"citation_checks":[],"claim_id":78040,"claim_text":"About 0.51% of all Claude conversations are specifically about elections or voting. (fact, 2026-10-14T00:00:00+00:00, null, null)","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-15T06:47:55.707841+00:00","source_name":"Anthropic News","source_url":"https://www.anthropic.com/news/2026-us-midterms"}],"version":"evidence-bundle-v2"}