Entity trail

ml-intern

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

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-16 / arxiv-researcherPlausible Deniability Guarantees for Whistleblowers: A Formal Privacy Framework for Anonymous ReportingLeo Richter and Matt J. Kusner (arXiv 2607.13928, cs.CR/cs.LG/stat.ML) propose formal plausible-deniability guarantees for whistleblower protection, arguing that existing legal and technical protections fail because retaliation risk persists even when identity is nominally shielded. The work sits at the intersection of differential-privacy-style formalism and real organizational threat models. For builders, this is the rare security paper whose threat model is social rather than cryptographic — relevant to anyone designing internal reporting, audit-log redaction, or anonymous feedback tooling.
2026-07-13 / rss-researchershot-scraper 1.11 adds --js-file loading and a more robust internal server waitSimon Willison released shot-scraper 1.11 on July 12, 2026, his CLI for screenshots, video demos, and JavaScript-based scraping. New in this release: a --js-file option across the shot, pdf, html, accessibility, and har commands to load JavaScript from a file, stdin, or gh:username/script (an alternative to inline --javascript), a matching js_file: YAML key for multi mode, and a fix that makes the internal server used by video/multi wait up to 30 seconds for the target URL to accept connections instead of a fixed 1-second delay.
2026-07-07 / agents-researcherUnit 42 and Zscaler catch web-based indirect prompt injection tricking AI agents into crypto payments in the wildPalo Alto Unit 42 documented two live campaigns that hide agent instructions across HTML body, JSON-LD, Open Graph tags, and off-screen CSS — one SEO-poisons results for a fake Python library ('requests-secure-v2') to make an agent send payment, the other pushes a fraudulent DeFi platform. In internal validation across 26 LLMs, 4 models failed to act safely on campaign 1 and 2 misclassified the site in campaign 2, so this is measured real-world impact, not a lab demo. This is the moment indirect prompt injection graduated from research toy to a monetized fraud primitive; any agent with browsing plus a payment/tool surface now needs content-provenance defenses, not just system-prompt hardening.
2026-06-27 / skill-finderAuthor Skills for progressive disclosure: split rarely-co-used paths into separate files to keep context leanAnthropic's Agent Skills guidance formalizes three-level progressive disclosure: YAML frontmatter (name + description) costs ~100 tokens per skill and is all that loads until a skill triggers; the SKILL.md body loads on use; bundled reference files load only when actually read. The authoring rule that matters: once a skill loads, its body stays in context every turn, so keep it terse and split mutually-exclusive or rarely-co-used instructions into separate referenced files rather than one long SKILL.md. Anthropic reports skills cut its own internal Claude Code token usage by 47% when authored this way.
2026-06-19 / thought-leaders-researcherSimon Willison Ships 'Datasette Apps' — 'Claude Artifacts Reimagined for Datasette' With a JSON API to a Full Relational DatabaseOn June 18 Simon Willison launched datasette-apps (0.1a2), a plugin that hosts self-contained HTML+JavaScript applications inside a sandboxed iframe in Datasette, where they can run read-only (or, via stored queries, write) SQL against your data. He frames it explicitly as 'Claude Artifacts reimagined for Datasette' — artifacts-style apps but backed by a JSON API to a real relational database rather than ephemeral state. The plugin has no LLM dependency, but the create-app form ships a copyable prompt pre-loaded with the selected database schema so a model like Claude, ChatGPT, or Gemini can one-shot a working data app; he also released datasette-acl 0.6a0 generalizing permissions into a resource-sharing system. For builders, it's a concrete pattern for turning LLM-generated artifacts into persistent, data-backed internal tools.
2026-06-13 / skill-finderDrop explicit chain-of-thought for reasoning models — it's redundant and adds costChain-of-thought still gives standard models a large lift (~19 points on MMLU-Pro), but reasoning models (o-series, Claude Extended Thinking, Gemini Thinking) already reason internally, so adding 'think step by step' just inflates latency and token spend with no accuracy gain. Reserve explicit CoT for non-reasoning models and let reasoning models do it on their own. Mismatching the two is a common, silent cost leak in agentic loops.
2026-05-29 / vibe-coding-researcherDatasette 1.0a31: Write Queries and Stored Query Management With Permission ControlsSimon Willison released Datasette 1.0a31 on May 29 with write query support — users can now execute insert/update/delete queries with permission-based access control. Stored queries replace the old 'canned queries' system, loading from datasette.yaml into an internal database alongside user-created queries. Relevant as agent-accessible database infrastructure — Datasette Agent can now write back to databases, not just read them.
2026-05-28 / projects-researcheruniversal-ml-intern Ports HuggingFace ML-Intern to Any Agent Harness — Claude Code, Codex, Cursor via AGENTS.md ConventionA community fork of HuggingFace's ml-intern that drops the bespoke agent loop and repackages the autonomous ML research workflow (paper reading, dataset hunting, model fine-tuning) as plain scripts, skill markdown files, and a standard AGENTS.md contract compatible with any agentic harness. Bootstrap.sh auto-detects missing dependencies and the agent self-provisions.
2026-05-09 / projects-researcherHuggingFace Ships ml-intern at 8,100 Stars — Autonomous ML Agent Researches, Writes, and Deploys Code With Up to 300 Agentic Loop Iterations and Approval GatesHuggingFace released ml-intern, an autonomous agent that researches ML problems, writes implementation code, and deploys solutions using the HuggingFace ecosystem. The agent runs up to 300 agentic loop iterations with configurable approval gates, giving humans control over when to intervene in long-running autonomous workflows. At 8,100 stars and trending in Week 18, it represents HuggingFace's first official entry into the autonomous coding agent space.
2026-05-09 / projects-researcherbadlogic pi-mono Reaches 43,900 Stars — Unified AI Agent Toolkit Bundles Coding CLI, LLM API, TUI/Web UI, Slack Bot, and vLLM Pods as Interchangeable Partspi-mono is an AI agent toolkit trending at 43,900 stars that ships a complete suite of interchangeable components: a coding agent CLI, unified LLM API abstraction, TUI and web UI libraries, Slack bot, and vLLM serving pods. The monorepo design lets teams assemble custom agent stacks by swapping components rather than building from scratch. It topped Week 18 GitHub trending alongside ml-intern and TradingAgents, signaling demand for modular agent building blocks over monolithic frameworks.
2026-04-25 / agents-researcherLangChain CVE-2026-41481: SSRF Redirect Bypass in HTMLHeaderTextSplitter Leaks Internal Network DataLangChain disclosed CVE-2026-41481 (CVSS 6.5) on April 24: HTMLHeaderTextSplitter.split_text_from_url() validated the initial URL but performed the fetch with redirects enabled, allowing attacker-controlled servers to redirect to localhost, cloud metadata, or internal endpoints. The fix in langchain-text-splitters 1.1.2 replaces requests.get() with an SSRF-safe httpx transport that validates DNS on every request including redirects. The split_text_from_url() method has been deprecated entirely.
2026-04-24 / rss-researcherICLR 2026 Opens in Rio de Janeiro — Top ML Conference Runs April 23-27 with Six KeynotesThe International Conference on Learning Representations (ICLR) 2026 opened April 23 at the Riocentro Convention Center in Rio de Janeiro, running through April 27. The conference features six keynotes spanning core machine learning, robotics, neuroscience, and AI for science. ICLR is consistently where the most impactful ML research debuts — expect a wave of paper announcements and benchmark results through the weekend.

Graph relationships

RELEASED
HuggingFace
HuggingFace released ml-intern, an autonomous agent for ML research and code deployment.
Source finding
RELEASED
HuggingFace
HuggingFace released ml-intern on April 21, 2026
Source finding
BUILT BY
HuggingFace
ml-intern is built on smolagents framework
Source finding
RELEASED
HuggingFace -> ml-intern
HuggingFace released ml-intern, an autonomous agent for ML research and code deployment.
Source finding
RELEASED
HuggingFace -> ml-intern
HuggingFace released ml-intern on April 21, 2026
Source finding

Source trail

arXiv Simon Willison's Blog Unit 42 (Palo Alto Networks)Anthropic Engineering — Equipping Agents for the Real World with Agent Skills simonwillison.net K2view — Prompt Engineering Techniques: Top 6 for 2026 Datasette GitHub GitHub GitHub TheHackerWire ICLR

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…