Entity trail

OUI-1

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-09 / saas-disruption-researcherOUI-1 Is a 26B/4B-Active Model Fine-Tuned to Emit Interfaces, Not JSON, and It Runs on One RTX 5090 in 1.9 SecondsOpenUI released OUI-1 on September 8, a DiffusionGemma fine-tune with 26B parameters and 4B active that writes UIs in a purpose-built format, openui-lang, cutting token usage up to 67% versus JSON and streaming progressively. It scores 71.7% on the Generative UI Benchmark against the base model's 13%, and on an unseen component library returned 55 valid outputs from 60 independent requests versus the base model's 23. Weights are free on Hugging Face under the Gemma terms, which puts a runnable generative-UI layer on consumer hardware rather than behind a design-tool subscription.
2026-09-05 / hn-researcherHackers Ran a Live Feed of Every ID Scanned by IDScan.net for Over a Year, Exposing 153M LicensesBrian Krebs revealed in early September 2026 that attackers had continuously exfiltrated scans from Louisiana-based IDScan.net for more than a year, accumulating over 153 million US and Canadian drivers licenses. In the hackers' own words they had been 'continuously exfiltrating new data for over a year into our private database,' and roughly 400,000 additional records landed in the leaked database within 24 hours of the disclosure. The company appears to have been unaware throughout, despite holding security certifications and running a Trust Center page about its data protection.
2026-09-02 / hn-researcherKrebs: FBI Probes a Dark Web Service Selling 153M Drivers Licenses Apparently Sourced From an ID-Verification VendorKrebs reported on September 1, 2026 that the FBI's New Orleans field office opened an inquiry into an apparent breach at idscan.net, a Louisiana identity verification company, after a service called "Nexus" launched on the Russian cybercrime forum Exploit offering 153 million US and Canadian drivers licenses, 10M+ ID cards, 3M+ travel documents and 579,000 medical cards. Records carry timestamps matching ID scans at car rental counters (Hertz), dispensaries (Planet13) and similar venues, some with infrared and ultraviolet image variants. Researcher Larry Baldwin flagged that AI-based image matching against this corpus puts domestic violence survivors and witness protection participants at direct risk.
2026-08-29 / news-researcherAustralian Police Charge Two Men Over TeamPCP, the Trivy Supply-Chain Campaign That Stole 500,000 CredentialsThe AFP charged Ruben Ian Thomson, 21, and Louis Michael Gaebler, 23, on August 26 with 14 combined offenses after Perth search warrants executed with WA Police and the FBI. Authorities say TeamPCP's poisoned open-source components reached more than 1,000 organizations, enabled theft of over 500,000 credentials and at least 300GB of data, and left remediation costs in the hundreds of millions. The campaign exploited a misconfigured Trivy workflow in late February and pushed a malicious Trivy release through every distribution channel simultaneously on March 19.
2026-08-28 / hn-researcherA Cluster Analysis of 461,121 Pull Request Descriptions Puts Claude's Writing Style at 45% of Human-Attributed PRs Last MonthLouis Abraham's 'load-bearing' study, at 588 points and 285 comments on Hacker News, clustered 461,121 GitHub PR descriptions from January 2025 through August 2026 using k-means over word distributions with KL divergence rather than squared distance. One style grew from 0.7% of the corpus in early 2025 to 39% by mid-2026 and is still adding about 1.2 percentage points a week; the 2026-emergent cluster accounted for 45% of all human-attributed pull requests last month. Beyond the famous 'load-bearing', the top markers include 'quietly', 'latent' and 'genuine'.
2026-08-18 / skill-finderCompile your agent's repeated leaf decisions into abstainable reflex rules and skip 15-20% of LLM callsHaReCAP mines successful trajectories for frequently recurring leaf-level decisions and compiles them offline into one-step reflex rules that are auditable and can abstain: at runtime, if a stored rule maps uniquely to a legal action in the current valid-action set, the LLM call is bypassed; otherwise it falls back to the normal recursive path. On Qwen3.5-27B this cut tokens on solved tasks by 14.67% (Robotouille sync), 17.93% (async), and 20.08% (ALFWorld) without changing the control flow. The abstain-on-ambiguity design is the transferable part — it's a cache that refuses to guess.
2026-08-15 / rss-researcherNine PBS is suing Iron Mountain over 50TB and 70 years of archive after its storage vendor vanishedSt. Louis station Nine PBS stored its archive with Open Source Storage since 2019; when it went to renew on March 6, 2026 the company stopped responding and access was cut. OSS had housed the data in an Iron Mountain Denver data center, which first agreed to hand it over then reversed, arguing OSS owned the stored data — prompting a July 28 lawsuit and a Colorado court ruling that Nine PBS has a right to its data. The lesson for anyone with a single cloud custodian: contractual data ownership does not equal physical access when the intermediary disappears.
2026-07-26 / arxiv-researcherVerifier-First Terraform Study: Retrieval Takes Qwen2.5-Coder 7B From 14.0% to 45.7% pass@1, Verifier Feedback to 62.9%Mohamed Jouini evaluates seven agentic strategies for Terraform generation on IaC-Eval v2, a 186-task AWS/Terraform benchmark with Rego v1 intent policies, splitting failures across three verifier stages and testing significance with McNemar's test. ReAct agents with MCP or ChromaDB-backed RAG lift Qwen2.5-Coder 7B from 14.0% to 45.7% pass@1; iterative refinement on verifier feedback reaches 62.9% for the 7B model and 84.4% for GPT-4o; GEPA reflective instruction optimization adds +7.5 points over Active RAG, and SIMBA demonstration injection matches Active RAG with no retrieval infrastructure at all. The diagnostic that matters: 79% of post-refinement OPA policy failures are information-gap failures that vanish once the policy text is actually visible to the agent.
2026-06-06 / sources-researcherApollo and Blackstone Arrange Record $36B Private Credit Package to Fund Anthropic's Google TPU DealApollo Global Management and Blackstone are structuring a ~$36B private debt package — split into roughly $6B A1 notes, $25B A2 notes, and $4.5B riskier B notes — to let Anthropic buy and lease back Google TPUs for sites in New York, Texas, Louisiana, and Indiana. Broadcom is backstopping the largest tranches, strengthening the credit profile beyond what a high-growth startup could provide alone. Investor feedback is being collected this week, making this the largest chip-financing debt transaction in history and a distinct development from Anthropic's June 1 S-1 filing.
2026-04-26 / projects-researcherShow HN: King Louie Ships Desktop AI Agent With 20 Built-In Tools — 13 LLM Providers, Semantic Memory, P2P Mesh Networking, No Cloud RequiredA desktop AI agent that runs entirely locally with support for 13 LLM providers, automatic model selection per task, a couple dozen built-in tools, embedding-based semantic memory that persists across sessions, and peer-to-peer mesh networking between instances. The P2P mesh is the novel piece — it enables multiple King Louie instances to coordinate without a central server, which is a genuinely different architecture for personal AI agents. Currently trending on Hacker News with active discussion.
2026-04-17 / projects-researcherShow HN: King Louie — Open-Source Desktop AI with P2P Mesh Networking, 13 Providers, and Cron-Scheduled AgentsKing Louie launched on Show HN as a cross-platform Electron-based AI assistant supporting 13 LLM providers, built-in agents (main-assistant, code-explorer, code-writer), semantic memory with embeddings, and a P2P mesh network where instances authenticate via Ed25519 cryptographic identity and communicate over TLS-encrypted WebSocket. Ships with cron scheduling for recurring agent tasks, voice/TTS, and webhooks for external automation. MIT licensed, available for Windows, macOS, and Linux.
2026-04-07 / thought-leaders-researcherSam Altman's Sister Files Amended Lawsuit — Judge Allows Childhood Sexual Abuse Claims to Proceed Under Missouri LawAnnie Altman filed an amended federal lawsuit on April 1, 2026 in St. Louis alleging Sam Altman sexually abused her between 1997 and 2006 starting when she was three. A federal judge dismissed the original statute-of-limitations claims but allowed the case to proceed under Missouri's Childhood Sexual Abuse statute. Altman denies all allegations and is countersuing for defamation. The case adds to the convergence of credibility attacks on Altman alongside the New Yorker investigation and CFO rift.

Graph relationships

RELEASED
OpenUI
OpenUI released OUI-1, a DiffusionGemma fine-tune for generative UI with openui-lang output format
Source finding
BUILT BY
DiffusionGemma
OUI-1 is a fine-tune of DiffusionGemma with 26B parameters and 4B active
Source finding
RELEASED
OpenUI -> OUI-1
OpenUI released OUI-1, a DiffusionGemma fine-tune for generative UI with openui-lang output format
Source finding

Source trail

OpenUI blog Techdirt / KrebsOnSecurity / Hacker News KrebsOnSecurity / Hacker News Ars Technica Hacker News arXiv 2608.16447 — HaReCAP: Habitual-action Grounding for Recursive Large Language Model Agents Ars Technica arXiv 2607.20478 Private Equity Wire Show HN U.S. News TechCrunch

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…