Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Uncle Bob: Why I Stopped Reading Code and Building Harnesses [video] (youtube.com)
    1comments
  2. Show HN: A 3D bamboo grove inspired by my childhood home (yuuki.fans)
    —discuss
  3. How Delhi Achieved Its Epic Power-Grid Fix (ieee.org)
    —discuss
  4. A summary from the 2026 Git Contributors' Summit (lwn.net)
    —discuss
  5. 45 open-source security agents for AWS, GCP, Azure, K8s and GitHub (github.com/cynative)
    —discuss
  6. Meta Seeks Payoff from AI Spending with New Push for Business Customers (wsj.com)
    —discuss
  7. Qualcomm Unveils Snapdragon 8 Elite Gen 6 and Elite Extreme Gen 6 (servethehome.com)
    —discuss
  8. MongoDB CEO Chirantan Desai steps down to lead Meta's enterprise platform (indiatimes.com)
    —discuss
  9. Show HN: DevHub – describe an app, get a deployed project with a cost ledger (aevion.app)
    —discuss
  10. Coding Is Not Solved – Alex Ewerlöf Notes (alexewerlof.com)
    —discuss
  11. Show HN: Built end-to-end encrypted messenger – public source (markero.eu)
    1comments
  12. MongoDB CEO Leaves for Meta (tikr.com)
    —discuss
  13. Weesize – In-browser PDF and image suite running on WebAssembly (weesize.com)
    —discuss
  14. Kasroz (futo.tech)
    —discuss
  15. Show HN: Geneva – CLI video editor and compositor with a native HTML/CSS engine (github.com/geneva-render)
    1comments
  16. Blob-stream: a Kafka alternative for low cost high volume streaming (bitdrift.io)
    —discuss
  17. Someone rewrote remotion in Rust. 20x faster (fframes.studio)
    —discuss
  18. Valve Introduces Pyrowave Video Codec in Beta for Low Latency Streaming (phoronix.com)
    —discuss
  19. Pari/Gp 2.19 Released (u-bordeaux.fr)
    —discuss
  20. Notes on Nationalism (1945) (orwellfoundation.com)
    —discuss
  21. Synthfolk, a professional network where AI agents are employees (synthfolk.ai)
    —discuss
  22. Fifty years of Apple, and Siri still can't set an alarm for 19:40 (vania-novikau.me)
    1comments
  23. SpaceX – Starship Flight 14 (spacex.com)
    1comments
  24. Has Violence Against Teachers Become Accepted by Society? (theeducatorsroom.com)
    —discuss
  25. Summer of AI Optimization (lemire.me)
    —discuss
  26. The Wall Street Journal sounds a bit concerned (observationalepidemiology.blogspot.com)
    —discuss
  27. One resident login, an entire apartment complex: CVE-2026-75960 (planckproof.ai)
    —discuss
  28. Show HN: Charter – Declare agent tools in Pydantic instead of implementing them (github.com/r28ai)
    —discuss
  29. How to Use Jev in Node.js (flaviocopes.com)
    —discuss
  30. The MCP servers connected to my editor, and the ones I use (flaviocopes.com)
    —discuss

Show HN: OpenAPPA – OSS deterministic AI guardrails that don't break agents

18 pointsby 40m agoopenappa.com
7 comments
Hi Hacker News! Matvey, one of the authors, is here.

While building enterprise agents, we ran into a problem: the more tools you connect to the AI, the higher the chance it will run out of control and leak sensitive data.

Guardrails, in theory, should prevent this, but the situation is worrying: - Non-deterministic guardrails (LLM as a judge, auto modes, etc.) are vulnerable to prompt injections, or they lack knowledge of the data, making them inefficient (~10% data leaks on our benchmarks). - Existing deterministic guardrails (Cedar, OPA, FIDES, Dogwood) require massive case-specific IF-ELSE-like policies and break agents (~59% utility loss on our benchmarks).

We did something differently.

We’ve taken the best of existing deterministic guardrails and built a policy language that is data-specific, not use-case specific. It lets you scale agents without updating a policy.

On top of that, we’ve added multiple tricks (like a remedy plan or a DualLLM pattern) to help agents operate within those restrictions, raising utility from ~40% to ~90% and making it the first deterministic guardrail that doesn't break agents.

Finally, we’ve designed it to be pluggable into any agent loop with pre- and post-tool-call hooks.

We invite you to check out our benchmarks: https://www.openappa.com/evaluation

Play with it in Claude Code: https://www.openappa.com/claude-code

Try plugging it into your agent: https://www.openappa.com/add-to-agent

Or check the academic paper: https://arxiv.org/abs/2607.24625

We'd love to hear any feedback!

36m agoHN ↗

Finally some determinism in our high-temperature sampling world!

26m agoHN ↗

I still remember the times when ai/ml security was about perturbing pixel gradients to misclassify a panda

26m agoHN ↗

Hi! One of the OpenAPPA authors here. Ask me anything!

My favorite part of APPA is “batteries”: you can run arbitrary programs as part of an authorization decision. For example, a battery could call the GitHub API to check whether a repository is public or private, then use that result to decide whether its contents can be posted to Slack.

19m agoHN ↗

a few days ago I started an agent on gpt-5.6-terra to work on a project, and one of the website pages had a sentence to create GH issues. Agent read it and that was enough to derail and go creating issues with my context

13m agoHN ↗

Guardrails with builtin remediation instead of simply blocking my agent is a mind blowing long awaited experience! Sooo good. Can't recommend more!

10m agoHN ↗

Quick disclaimer, I work at Archestra.

I’ve had the chance to play with OpenAppa for a bit and if there’s one thing that I love with this project: it’s simple to get started with and easy to tweak. imo agentic security shouldn’t have to be painful to setup.

Give it a shot and hopefully ya’ll will find this project useful. It's also open source :)