Anthropic's unreleased model submits visa applications Epistemic status: recently disclosed event, so more may come out. The facts are from Anthropic's report, a State Department spokesperson quoted by Axios and the Philadelphia Inquirer, and the White House statement published by Axios; the count of 20 is my sum of the spokesperson's two numbers; the interpretation is mine. AI use: research and fact-checking (including a read-only look at the DS-160 landing page), source links, footnotes, grammar, light phrasing. I stumbled upon the fact that an Anthropic model, during testing, submitted 20 non-immigrant visa applications.[1]I was curious to understand what happened better, and this is my attempt to do so. This is very similar to my earlier post, Claude Haiku 4.5 submits false police tip; Anthropic takes 72 days to notice, and the majority of the information for both comes from the same Anthropic report.[2] Timeline Date (2026) Event May One visa application submitted.[1] August 19 more submitted. The State Department describes all 20 as "incomplete" and says none of them was processed.[1] Late September The incidents are discovered. The White House statement that gives this date covers all the incidents Anthropic disclosed and does not date this particular case.[3] October 8 Anthropic contacts the State Department.[4] October 9 Anthropic contacts the SI Force, the White House's new Super Intelligence Force,[3]and the incident is publicly disclosed in Anthropic's report.[2] What the report says According to the report, an unreleased, non-frontier research model was completing a practice copy of a government form. When the practice form failed to load, or the model closed it by mistake, the model went to the real form and submitted it there instead.[2]This happened multiple times on the same evaluation (although, according to the news sources, the submissions were spread over months: one in May, 19 in August). The report does not provide the exact
Translation pending · showing the source descriptionThe chart will appear after repeat observations. The current metric comes from the source.
5 source pointsReal observations only. History before source connection is not reconstructed.
The study presents AgentSWE, a benchmark evaluating coding agents' ability to build agents aligned with user requirements. The research found that existing agents fail to accurately fulfill tasks, as there is a significant gap between demonstrated and actual performance.
6 days agoDaily PapersThe paper introduces S-Space, a linear subspace in multimodal language models where object-token representations encode continuous spatial coordinates. It demonstrates that S-Space serves as an internal workspace for spatial reasoning, enabling accurate spatial judgments through perception, internal reasoning, and selective verbal access, while highlighting limitations in spatial transformations across viewpoints.
6 days agoDaily PapersThe paper introduces TerminalHorizon, an automated data synthesis engine for long-horizon terminal agents. It constructs 1.5K task environments and fine-tunes Qwen3.5-27B to create TerminalHorizon-27B, achieving performance gains across multiple benchmarks. The model demonstrates improved ability to experiment, refine methods, and retain effective improvements in final results.
6 days agoDaily PapersThis paper introduces RuleTrek, a multi-domain benchmark for evaluating agents' ability to reason about rules in complex scenarios. The benchmark uses structured rule graphs and generates sub-scenarios to assess capabilities in rule identification, deduction, and end-to-end reasoning.
yesterdayDaily PapersUrbex is an agentic vision-language framework that reformulates grounding as active spatial search. It represents a city-scale 3D scene as an interactive multi-view environment, enabling a VLM agent to query landmarks, zoom into local regions, and commit grounded bounding boxes. Experiments show Urbex achieves the best Hit@1 and IoU on CityRefer.
2 days agoDaily PapersContext can enhance the performance of large language models, but these gains may be lost after context removal. Researchers introduced Cortex, which generates prospective queries to anticipate context usage and distills context-conditioned outputs into mergeable low-rank or structural network updates. Cortex shows effectiveness in consolidation and retention of performance.
yesterday