Self-hosted AI security testing with a local Web UI, live agent telemetry, verified findings, and branded PDF reports. Comments URL: https://news.ycombinator.com/item?id=48805464 Points: 3 # Comments: 0
Article URL: https://blog.brendankeaton.com/self-service-ransomware-as-security-against-local-ai-tools Comments URL: https://news.ycombinator.com/item?id=48805363 Points: 2 # Comments: 0
hey guys, i wanted to show one of my side projects. The idea is a coding harness (independent of models) natively designed for C/C++ developer workflows. I'm a C++ dev and do not find claude code work well with C++ toolchain like gdb and perf. The current version has integrations for gdb, clang-tidy, cppcheck, sanitizers, perf, benchmarking, compile DB navigation, Godbolt, symbolization, binary inspection, and decompilation. It supports Anthropic, OpenAI, Gemini, and self-hosted models. There ar
<!-- SC_OFF --><div class="md"><p>Built a memory system called TRACE that organizes agent conversation history into a topic tree (branches + summaries) instead of flat RAG chunks, and benchmarked it on MemoryAgentBench (ICLR 2026), specifically the EventQA accurate-retrieval task.</p> <p>Its a pypi package:</p> <p>pip install trace-memory</p> <p>Results (F1):<br/> • TRACE (gpt-oss-20B): 82.5%<br/> • TRACE (gpt-oss-120B): 83.8%<br/> • Mem0 (GPT-4o-mini, paper’s official number): 37.5%<br/> • MemG
Article URL: https://github.com/shenli/awesome-agent-infra Comments URL: https://news.ycombinator.com/item?id=48805260 Points: 3 # Comments: 0
When checking that solutions to certain problems are correct, it turns out, you can’t get around the inherent complexity of the quantum world. The post Researchers Reveal the Power of ‘Quantum Proofs’ first appeared on Quanta Magazine
We’ve got another special episode of Decoder today, recorded at the Cannes Lions advertising festival in the South of France. I’m talking with Ali Berman and Raina Penchansky, who run the Creators division at United Talent Agency. UTA is an enormous talent agency. Half the people you’ve ever heard speak or perform or who show […]
Claude Design is typically used to create prototypes, but not many people know that it’s possible to use this tool to craft very… Continue reading on UX Planet »
Article URL: https://fortune.com/article/ai-communication-undermining-human-relationships-middle-management/ Comments URL: https://news.ycombinator.com/item?id=48805032 Points: 1 # Comments: 1
Article URL: https://aisle.com/blog/mythos-at-home-and-its-called-aisle Comments URL: https://news.ycombinator.com/item?id=48805007 Points: 1 # Comments: 0
Article URL: https://sfconservancy.org/blog/2026/jul/03/github-gen-ai-california-22757-ok-for-foss-license/ Comments URL: https://news.ycombinator.com/item?id=48804983 Points: 1 # Comments: 1
FCA official makes case for greater powers for watchdog as millions use technology for personal finance decisions.
A controlled comparison of ungated vs gated AI patching on SWE-bench Lite. Full methodology, results, and reproducible scripts in the repo. Breakthrough for flash models. Comments URL: https://news.ycombinator.com/item?id=48804907 Points: 1 # Comments: 0
Article URL: https://github.com/CURSED-ME/loopers-oss Comments URL: https://news.ycombinator.com/item?id=48804865 Points: 1 # Comments: 0
Article URL: http://www.techdirt.com/2026/07/06/metas-ai-perv-glasses-now-come-with-stupid-comcast-esque-usage-restrictions/ Comments URL: https://news.ycombinator.com/item?id=48804836 Points: 6 # Comments: 2
You describe the model. It writes the code, runs the training, and ships the checkpoint. Welcome to ML Intern.
Article URL: https://github.com/Lore-Hex/BurstyRouter Comments URL: https://news.ycombinator.com/item?id=48804448 Points: 1 # Comments: 0
<pre style='white-space:pre-wrap;width:81ex'>refactor(agents): localize internal-only symbols (#100909)</pre>
Samsung is about to post a huge profit increase estimate, analysts say.
<pre style='white-space:pre-wrap;width:81ex'>fix(agents): preserve blank lines in multi-question user input fallback parsing (#100832)</pre>
Article URL: https://blog.mozilla.ai/introducing-otari-the-open-source-llm-control-plane/ Comments URL: https://news.ycombinator.com/item?id=48804378 Points: 3 # Comments: 0
Enterprise Document Intelligence [Vol.1 #8C] - Structured output is the start of validation, not the end: check the evidence, accept not-found, loop the feedback The post Validating the RAG Answer Before the User Sees It: Spans, Quotes, and the Feedback Loop appeared first on Towards Data Science .
<p> An AI coworker that lives in your Slack to get work done </p> <p> <a href="https://www.producthunt.com/products/yasmine-works?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1189393?app_id=339">Link</a> </p>
<p>If you told me back in 2014 that an idea for helping CD collectors would eventually become AI infrastructure, I probably wouldn't have believed you.</p> <p>But looking back, the path actually makes sense.</p> <p><em>Also, fair warning: this is a pretty personal story. It’s less about features and more about how an idea slowly evolved over time.</em></p> <h2> It started with a duplicate CD </h2> <p>Back in 2014 I was visiting a friend.</p> <p>His father is a serious CD collector. Thousands of
RoboBusiness 2026, the premier event for developers of commercial robotics and those building robotics businesses, is seeking expert speakers. The post RoboBusiness 2026 call for speakers closes soon appeared first on The Robot Report .
Hi HN! I built Belgie, a Python library that embeds Deno, allowing Python applications to run JavaScript and TypeScript without requiring Node.js or Deno to be installed on the host system. It supports the features you’d expect from Deno (npm, JSR, and URL imports, isolated package environments, etc.), plus: - Pass JSON-safe values seamlessly between Python and JavaScript. - Agentic code generation with typed interfaces, similar to Pydantic’s Monty. - Manage dependencies programmatically or from
<pre style='white-space:pre-wrap;width:81ex'>fix(feishu): send card JSON message params as cards (#100883) * fix(feishu): send plain card JSON as interactive cards Co-authored-by: martingarramon <263922628+martingarramon@users.noreply.github.com> * fix(clownfish): address review for gitcrawl-21-autonomous-drip-20260706 (1) Co-authored-by: martingarramon <263922628+martingarramon@users.noreply.github.com> * fix(feishu): preserve native card compatibility Reported-by: @ZenoRewn Co-authored-by: mar
<h1> Claude Code's China Detector Is the Wrong Kind of Security Control </h1> <p>Alibaba reportedly told employees to stop using Claude Code at work from July 10 after the tool was flagged for China-linked user detection code. Reuters framed it as a workplace ban over alleged backdoor risk. TechCrunch reported Anthropic's explanation too: Thariq Shihipar said it was an experiment launched in March to prevent account abuse by unauthorized resellers and protect against model distillation, and that
<pre style='white-space:pre-wrap;width:81ex'>fix(anthropic): restore subscription OAuth billing routing</pre>
Article URL: https://github.com/makerchecker/MakerChecker Comments URL: https://news.ycombinator.com/item?id=48804182 Points: 11 # Comments: 5
Article URL: https://www.appblit.com/pdfreflow/demo Comments URL: https://news.ycombinator.com/item?id=48804162 Points: 4 # Comments: 0
I made a strategy game where you play the US or China through the AI race, 2026 to 2030, sixteen quarterly turns in the browser. One run takes about half an hour. At the start, the game seals two dice you never get to see. Inside: how hard alignment really is, and how fast takeoff compounds. You get eval reports, but only as ranges, and they flatter you most exactly when your systems are least aligned. At the end you get a debrief which shows what your evals said each quarter and also what was a
MicroPython (lexer, compiler and VM) on the SNES: 3.58 MHz 65816, 56 KB Python heap, 16-bit int. The REPL runs right inside the post via EmulatorJS and also works on real hardware via flashcart. I did this as a benchmark for Claude Fable. When the export ban hit, switching to Opus got the project stuck for three weeks; Fable came back and found the real bug in ninety minutes. Along the way: 23 compiler bugs and 4 MicroPython bugs, each root-caused with a minimal reproducer and filed upstream. If
A streaming box should not need a threat model. Neither should a username field, a demo repo, a reset flow, or a browser permission prompt. That is the irritating part this week: the risky pieces were ordinary. Home devices became a routing cover. Clean code pulled dirt from a dependency. Identity shortcuts aged badly. AI systems trusted the wrong instructions. Same soft spot throughout: trust
Station F, a Paris-based startup hub founded by French billionaire Xavier Niel, is gearing up for a new edition of its F/ai accelerator program in a bid to strengthen its positioning as a stepping stone for promising AI startups.
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. South Korea’s hottest new bachelors are chip workers Baek, a 35-year-old manager at the South Korean semiconductor titan SK Hynix, was enrolled in a matchmaking company a year ago. In a…
Oracle's down more than 40% this month, the BIS thinks AI could destroy the economy, and we've got the Kettle on for a chat about the whole mess
Article URL: https://github.com/GrayboxTech/weightslab Comments URL: https://news.ycombinator.com/item?id=48803966 Points: 1 # Comments: 0
<pre style='white-space:pre-wrap;width:81ex'>fix(diagnostics-otel): route OTLP exports through env proxy * fix(diagnostics-otel): route OTLP exports through env proxy * fix(diagnostics-otel): harden OTLP proxy agent options * fix(ci): refresh rebased gateway checks * fix(ci): avoid proxy boundary scan</pre>
One thing I expected from the AI era was a decrease in emotional attachment to code reviews. My assumption was that AI-generated code would make reviews easier because criticism of the code wouldn't feel like criticism of the author. Instead, I've repeatedly seen people strongly defend AI generated code, and sometimes even defend AI explanations that are demonstrably incorrect. Is anyone else seeing this? If so, what do you think is driving it? How are you handling it within your team? Also, has
