frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Are we defaulting to VM-level sandboxing before understanding the threat model?

2•samhita-alla•1h ago
Hey everyone, I'm Samhita and I work at Union.ai. We've been building infrastructure for running agents and building models, which naturally got us thinking a lot about sandboxing. One thing I've been wondering: are we overusing heavyweight sandboxing solutions?

I think some form of isolation is non-negotiable when you're running model-generated code. Things like process isolation, filesystem restrictions, network controls etc. make complete sense. What I'm less sure about is whether VM-based approaches are necessary as often as people seem to think.

In my experience using coding agents locally, basic guardrails and sensible restrictions have been enough most of the time, at least when I'm operating in a relatively controlled environment and not deliberately pushing the agent into risky situations. Of course, that's very different from a production service running arbitrary user code.

So I'm curious: - What are you using to sandbox agents today? - What threat model are you optimizing for? - Have you had incidents that convinced you VM-level isolation was necessary? - Where do you draw the line between "good enough" and "needs stronger isolation"?

Would love to hear what has worked (or not worked) for others.

Comments

cpburns2009•49m ago
This is what I've been working on the past month: a strategy for running sandboxed containers. It's not strictly for agents (I'm now using it for OpenCode). You should be thinking about supply chain attacks for all of your applications that use third-party dependencies. PyPI and NPM have had a lot of compromised packages recently. The litellm hack affected a lot of agents, and there have been some Docker escape exploits.

The threat model I'm concerned about is supply chain attacks on third-party package repositories. The primary goal is to keep the convenience of containers, but to limit the blast radius of a compromised package or application, and reduce the risk of container escape. The software stack I'm currently evaluating is:

- Kata Containers: Backend for containerd to run each container in a KVM-backed QEMU microVM (alternative to the standard runc).

- containerd: Container runtime. Docker and Podman are not compatible with Kata 3. Kata 4 is supposed to fix that.

- nerdctl: Docker-compatible front-end to containerd.

- cni-plugins: networking component for containerd. Used to isolate containers networks.

- iron-proxy: MitM, TLS-intercepting egress proxy. This restricts all outbound traffic to whitelisted domains and IPs, and supports secret injection. Squid is a more established alternative.

How is this used in practice? I have a small bash script to launch the sandboxed OpenCode container with the current folder bind-mounted. OpenCode only has file-system access to the context directory, and limited network/internet access.

World Time, Date and Weather Resource

https://timeandcalendars.com:443/
1•FTU-Jose•26s ago•0 comments

Anguished Parents, Crying Doctors: Life Amid Utah's Measles Outbreak

https://www.wired.com/story/anguished-parents-crying-doctors-life-amid-utahs-measles-outbreak/
1•mooreds•41s ago•0 comments

Angels Landing trail in Zion National Park closed until further notice

https://www.sfgate.com/national-parks/article/angels-landing-zion-park-closed-22300369.php
1•mooreds•55s ago•0 comments

Control Planes Are Control Systems

https://byatt.io/posts/control-plane-as-control-system/
1•wmjbyatt•1m ago•0 comments

China prepares $295B plan to fund nationwide AI data center buildout

https://www.reuters.com/world/china/china-prepares-295-billion-plan-fund-nationwide-ai-buildout-b...
2•alephnerd•3m ago•0 comments

There Is(Ǝ) – Such That (∋)

https://www.fractalkitty.com/there-is-3-such-that/
1•evakhoury•4m ago•0 comments

PiLSMer: A data-free key-value store

https://norwood.github.io/PiLSMer/
1•gwen-shapira•5m ago•0 comments

The enterprise identity crisis: Who's Alice?

https://www.firezone.dev/blog/the-enterprise-identity-crisis-part-one
1•jamilbk•6m ago•0 comments

South Korea fines e-commerce giant Coupang $400M over data breach

https://www.bbc.com/news/articles/cvgj4rgz2n2o
2•ChrisArchitect•6m ago•1 comments

Show HN: Domain Rating – a leaderboard of startup website Domain Ratings

https://www.domain-rating.com
1•nocodeg•6m ago•0 comments

A gene regulates vertebrate growth, maturity, and lifespan

https://www.nature.com/articles/s41467-026-72381-0
1•wslh•7m ago•0 comments

The outsized impact of cultural idiosyncrasies

https://luke.hsiao.dev/blog/cultural-idiosyncrasies/
1•lwhsiao•7m ago•0 comments

AI and the Productivity Paradox

https://www.ft.com/content/b4b60d00-2e8c-4db0-b3ed-9988dc0eeb5c
2•1vuio0pswjnm7•7m ago•0 comments

Liebreich: The Great Clean Energy Acceleration 2.0 – BloombergNEF

https://about.bnef.com/insights/clean-energy/liebreich-the-great-clean-energy-acceleration-2-0/
1•xbmcuser•8m ago•0 comments

The $1M AWS Server [video]

https://www.youtube.com/shorts/tQ48PxXPd88
1•sunbirdLabs•8m ago•0 comments

Show HN: Brooks-Lint – AI code reviews grounded in 12 classic engineering books

https://github.com/hyhmrright/brooks-lint
1•hyhmrright•9m ago•0 comments

The 90-year-old idea behind JEPA models: Canonical Correlation Analysis

https://shonczinner.github.io/posts/embedding-prediction/
3•Anon84•10m ago•0 comments

Ask HN: Is anyone else seeing a Slack auth bug?

1•HoyaSaxa•10m ago•0 comments

Elon Musk Is About to Make Saving for Retirement Even Harder

https://prospect.org/2026/06/10/elon-musk-spacex-ipo-retirement-savings-index-funds-stock-market/
2•dxs•10m ago•0 comments

Explain AI: AI App Directory and Governance Platform

https://www.explain-ai.com
1•Gast•10m ago•0 comments

Euro-Office: First version of the open-source web office is here

https://www.heise.de/en/news/Euro-Office-First-version-of-the-open-source-web-office-is-here-1132...
2•doener•11m ago•1 comments

The Model Is No Longer the Bottleneck

https://www.k-dense.ai/blog/the-model-is-no-longer-the-bottleneck
2•gmays•11m ago•0 comments

The Impossible Shift

https://soon.works/the-impossible-shift
1•ma5ly•11m ago•0 comments

Show HN: A GPT-realtime-2 tool that navigates your site by voice

https://www.mascotlyai.com
1•travelingtice•12m ago•0 comments

MapComplete – Contibute to OpenStreetMaps

https://mapcomplete.org/
3•GTP•13m ago•0 comments

Helm AI Kernel, a fail-closed execution firewall for AI agents

https://github.com/Mindburn-Labs/helm-ai-kernel
2•mindburnlabs•14m ago•0 comments

Man sues Florida cops over arrest spurred by "93% match" in facial recognition

https://arstechnica.com/tech-policy/2026/06/man-jailed-due-to-faulty-face-recognition-says-florid...
7•Brajeshwar•15m ago•1 comments

Explosive Weapons Monitor 2025

https://explosiveweaponsmonitor.org/reports/9/explosive-weapons-monitor-2025/
1•lode•15m ago•0 comments

Which LLM is the best proofreader?

https://revise.io/errata-bench
1•artursapek•16m ago•0 comments

Show HN: Drawdown-protected custom FIRE portfolios with API for rebalance alerts

https://algorithmicfire.com/user/alerts
1•algorithmicfire•18m ago•0 comments