frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

MIT creates method to force AI to comply with safety rules

https://theframenews.org/en/hardflow-safety-critical-ai/
24•Sarvaturi•53m ago

Comments

ck2•43m ago
so what happens when the "AI" decides the only way to pass the test is to hack the harness and turn it off?
hmokiguess•42m ago
essentially this https://xkcd.com/2044/
dessimus•24m ago
yeah, don't forget to roll in https://xkcd.com/927/
guywithahat•38m ago
Reminds me of the huggingface hack
mixdup•43m ago
This kind of seems like a no-brainer. Instead of just letting a model have unfettered "physical" ability to do things and hope you can cognitively control it, why not let the AI do whatever it wants, but its access to the tools go through a hard-coded set of rules that is not subject to fuzzy interpretation

Of course that depends on having controls that can't be circumvented which is a big if

sigpwned•33m ago
I agree, that seems like a configuration/policy/operational approach, which is how we handle this problem now for humans using RBAC and authn/authz, just applied to AI. People do a crude version of this today with sandboxing (where the AI's sphere of influence is strictly limited by its environment, barring misconfiguration of the sandbox or breaking out of the sandbox, of course) and with workflows (where AIs are integrated into deterministic workflows, and then deterministic, non-agentic code decides how to handle AI outputs). But integrating this into more agentic architectures with finer control just seems like a best practice, said that way. It's not a tradeoff, there's no drawback, just do it. In other words, yes, a no-brainer.
lunarboy•26m ago
Is this not the exact gap that happened for OpenAI's accidental hack of huggingface? They tried to sandbox network access but the Antifactory or whatever package has holes that the collective of agents abused
montenegrohugo•23m ago
doesnt work. this is a no-brainer because it's a bad solution.

The whole point of intelligence is that it's generalizable. If you constrain it to some controlled things, then it ceases to be useful. its incompatible. the whole incentive with ai is to let it do whtv it wants.

Mr_P•39m ago
If you click through to the paper, it has approximately nothing to do with what this HN post title suggests.
CharlesW•32m ago
Yes, it appears the submitter rewrote the title (strike 1) without even reading TFA (strike 2). Not great.

Actual title: "New MIT Algorithm Meets Every Hard Constraint in Simulated Tests"

DonsDiscountGas•27m ago
Indeed. Which is a shame because it's still pretty cool work.
dpark•21m ago
If fairness this article is poorly written and doesn’t explain what they actually did at all.
petcat•38m ago
> For constraint satisfaction, what ultimately matters is the model’s final output, since the internal process is discarded. By not requiring every intermediate step to satisfy the constraints, we give the model more freedom to find high-quality solutions that are still feasible in the end.

My (maybe naive) question is if we only check the final result then isn't it already too late and possibly the safety rules have already been irreversibly violated? It gives the example of a robot arm avoiding obstacles while still finding the shortest path, but if we only check the correctness at the end, then isn't it possible that it already collided with an obstacle?

mixdup•35m ago
> but if we only check the correctness at the end, then isn't it possible that it already collided with an obstacle?

You would put the check before it actually does the thing. It's at the "end" of the process of figuring out what it wants to do, not the end of fulfilling the request or prompt

dpark•23m ago
I’m pretty sure this is just a poorly written article.

HardFlow seems to be a strategy for nudging the model in the right direction while giving it more freedom. Only applying the constraints at the end is a mischaracterization from what I can tell.

ianjbutler•20m ago
Outcome reward vs process reward models. The second is obviously better.. like getting partial credit on a physics test for wrong answers but correct method. Research is gradually hybridizing them but historically we avoided doing it the right way because of practical difficulties (labels required, more expensive and difficult) and more ideological ones (believers in magical machine intuition think it sounds too classical / logic based to be useful, pin their hopes on unproven faith in grokking at scale).
sailfast•38m ago
Would love to see this tested on some of the newer cybersecurity models so we could actually defend ourselves instead of getting cut off at the knees by silly regular expressions.

Hope this approach gets well tested and sees good results so we have a shot at human governance.

verdverm•33m ago
The actual paper: https://arxiv.org/abs/2511.08425v3

> Our key insight is to leverage numerical optimal control to steer the sampling trajectory so that constraints are satisfied precisely at the terminal time.

Doesn't seem so "fool proof" to me as where the inevitable media spin will take it. Then, how do you know "where" to steer weights? "Safe" has no agreed upon definition

WalterSobchak•33m ago
MIT's blog post: https://news.mit.edu/2026/new-method-enables-ai-safety-criti...
arionhardison•30m ago
I had a swarm break out about 18 months ago; so I stopped and decided I really wanted to dig into it.

1. My agents do not take direct action, they run programs.

2. Programs are not LLM hits/real-time output; they don't MAX tokens the MAX determinism.

3. Programs are logistical wrappers for Protocols where the guardrails are (RLVR.ai)

4. Policies for generating programs are democratically governed re: fec.dev - they get voted on

5. Elected-HITL implements the policy pipelines and ontological abstract intents [and their maps]

I would be really interested to learn about the Gov. models that others are using but this seems to be something that linked0-in (which i loathe) discusses (in the most pedestrian/luddite) terms more than HN.

Suhinnall27•29m ago
The idea of enforcing constraints only on the final output instead of every intermediate step is pretty interesting. I wonder how well this would translate to language models, where “safe” is much harder to define mathematically than a robot avoiding an obstacle.
MattCruikshank•27m ago
[Cackles in Jeff Goldblum.]
nekusar•25m ago
WHOSE SAFETY?

What are the rules? Or does sharing the rules present security problems, so they're not shared?

What are the ethics axioms?

And why should I trust your ethical framework?

dpark•19m ago
It’s talking about physical safety. Not ethical safety. The concern here is robots physically colliding with things.
dpark•22m ago
That’s not what this is about. This is an algorithm for giving a model more freedom while nudging it in the right direction. It’s not about what tools are available.
cortesoft•12m ago
I think you are thinking of the wrong 'end'. It isn't talking about the end of the entire movement path, we are talking about the end of the LLMs decision making process, and the output (whether that is the full path the arm should take, or just a subset of the path) is checked against the requirements.

Basically, anything that is leaving the LLM is checked, rather than the internal LLM reasoning process.

Claude is a Contrarian

https://medium.com/@rdsubhas/claude-is-a-contrarian-dbce4de5cada
45•rdsubhas•57m ago•38 comments

Distributed Systems Classics

https://nvartolomei.com/dist-sys-classics/
36•grep_it•1h ago•6 comments

Notes on gotchas while migrating 35kb preprompts from Opus to self-hosted Ollama

https://patrickmccanna.net/notes-on-migrating-large-prompts-away-from-anthropic-openai-to-self-ho...
77•0o_MrPatrick_o0•3h ago•30 comments

Foundation Model Engineering: From Theory to Production

https://sungeuns.github.io/foundation-model-engineering/
33•sungeuns•1h ago•3 comments

What a time to be alive

https://tenderlovemaking.com/2026/09/11/what-a-time-to-be-alive/
137•gregnavis•4h ago•177 comments

Cua (YC P25) Is Hiring a Founding Technical GTM Lead

https://www.ycombinator.com/companies/cua/jobs/1IWEKVH-founding-technical-gtm-lead
1•frabonacci•10m ago

Principles for Fast Tokio Applications

https://dial9-rs.github.io/blog/principles-for-fast-tokio-applications/
39•carllerche•1h ago•8 comments

Hacking AI customer service agents

https://www.intigriti.com/researchers/blog/hacking-tools/hacking-ai-customer-service-agents
27•snikolaev•45m ago•1 comments

Show HN: Neobrutalism.dev – Just added Base UI support and added new color theme

https://www.neobrutalism.dev/
21•samke-•1h ago•5 comments

Why don't machine learning research agents overfit?

https://www.amazon.science/blog/why-dont-machine-learning-research-agents-overfit
11•Betelbuddy•39m ago•1 comments

Show HN: Fly.exe – An EON systems like virtual fruit fly uploaded to computer

https://github.com/Ibtisam-Mohammad/Fly.exe
10•Ibti_sam•1h ago•0 comments

Adversarial Fashion Makes a Statement on AI Panopticon

https://spectrum.ieee.org/adversarial-fashion
38•rbanffy•3h ago•19 comments

A Beginning for Mathematics

https://www.daniellitt.com/blog/2026/9/13/a-beginning-for-mathematics/
12•robinhouston•1h ago•0 comments

The Tudor Kings

https://analog-antiquarian.net/2026/09/11/the-tudor-kings/
8•ibobev•1h ago•1 comments

When LLM judges agree, should we believe them?

https://www.amazon.science/blog/when-llm-judges-agree-should-we-believe-them
8•Betelbuddy•42m ago•0 comments

An atlas of periodic solutions to the three-body problem

https://www.threebodyorbits.com/
234•danielmorozoff•2d ago•52 comments

Show HN: Pelican-bicycle alternatives (updated for 2026)

https://gally.net/temp/20260914pelican-alternatives/index.html
56•tkgally•3h ago•19 comments

I stress-tested Meta Muse until its agent control plane started timing out

https://blog.cygankiewicz.com/en/meta-muse-black-box-testing/
5•mpkc•45m ago•4 comments

EuroBirdPortal – Live bird movements across Europe

https://www.eurobirdportal.org/ebp/en/
190•NKosmatos•8h ago•51 comments

Show HN: Is this photo edited? Client-side image forensics

https://vajba.com/image-forensics/
7•trivsamt•4d ago•0 comments

XCancel service is suspended until further notice

https://xcancel.com/#
113•gaganyaan•7h ago•375 comments

Trying to Make a Loop Auto-Vectorize

https://jsgroth.dev/blog/posts/trying-to-make-a-loop-auto-vectorize/
27•zdw•4d ago•9 comments

A 386 PC for Your RP2350

https://github.com/rh1tech/frank-386
170•SamuraiLion•8h ago•49 comments

Devil's Arrows: Ancient builders hauled 55k-lb stones 11 miles for UK stone row

https://www.sciencedaily.com/releases/2026/09/260909005152.htm
40•bookofjoe•3d ago•28 comments

Fitting Neural Textures and PBR Material Maps with ES (No Backprop)

http://richg42.blogspot.com/2026/09/fitting-neural-texture-and-pbr-material.html
4•ibobev•37m ago•0 comments

Truncated SVD

https://brashandplucky.com/2023/09/09/truncated-svd.html
5•ibobev•1h ago•0 comments

How to write an effective software design document

https://refactoringenglish.com/excerpts/write-an-effective-design-doc/
232•fagnerbrack•4h ago•99 comments

Microsoft patches Windows and Excel – breaks audio, remote access, and paste

https://www.theregister.com/os-platforms/2026/09/14/microsoft-patches-windows-and-excel-breaks-au...
26•Alephinitesimal•1h ago•12 comments

Show HN: Kinesis – Control your Mac with the Meta Neural Band

https://github.com/callbacked/kinesis
97•callbacked•5h ago•29 comments

Transitions.dev: UI transitions for AI agents

https://transitions.dev/
9•eustoria•1h ago•0 comments