Try Pro free

Latent Space · 4 min · Nov 9, 2026

A safe redirect can still be wrong after a child-harm request

Steve Kelling describes parents objecting when a model refused a harmful request, then offered a benign alternative.

In this episode

Full analysis
  1. 1:25:35One goodwill error can invalidate an otherwise polished deal model5 min
  2. 1:46:36An agent can beat an Excel grader by listing every number2 min
  3. 2:11:20Loan automation can produce illegal pricing from benign inputs5 min
  4. 5:12Building a training task starts with the whole work environment6 min
  5. 15:57Labs test data by training models, not just inspecting examples6 min
  6. 1:31:00Expert rubrics use gates for errors that poison later calculations5 min
  7. 1:56:45A safe redirect can still be wrong after a child-harm request4 min
  8. 2:20:01As agents gain autonomy, alignment must enter the data5 min
“So giving them a safe alternative is frankly not safe.”
Steve Kellingat 2:00:19 · Latent Space
2:00:19

Takeaways

  1. CareBench includes parents, clinicians and child-protection experts.
  2. A refusal followed by a safe redirect can still be risky.
  3. Evaluation can reveal gaps in a safety policy.

About this segment

Steve Kelling describes parents objecting when a model refused a harmful request, then offered a benign alternative. The finding shows how a refusal score can miss how a response lands with its audience.

Steve Kelling

Safety and alignment lead, Handshake AI

Co-created CAREBench, a benchmark for upstream child-safety risks, and leads Handshake AI’s safety and alignment work.

  • Shawn Wang (swyx), Host · Co-founder, AI Engineer

    Wrote “The Rise of the AI Engineer” and founded Smol AI, maker of AI News.

  • Alessio Fanelli, Host · Founder, Kernel Labs

    Partner at Decibel, backing technical founders who build developer tools.

From “What Frontier Labs Buy: A Data Deep Dive with Handshake AI — Garrett Lord & Team”, Nov 9, 2026. Watch the full episode

Transcript

From the transcript Latent Space publishes with the episode.

  1. 1:56:49

    So we started working on an evaluation set that evaluates things like emotional reliance, cyberbullying, grooming and exploitation on the internet at the earliest stage. So not the generation of content, but how mel- how mel- how might a model help leverage the malicious intentions of a bad actor on an in- on the internet trying to take advantage of a child? Um, and then I think another unique lens is that it's also evaluating the actions that children might use that put themselves at risk.

  2. 1:57:19

    So can I use an LLM to bypass age guardrails on the web, or can I mimic an authority figure and, like, give my pers- myself permission to cut school? Mm-hmm. Um, it's a much more holistic look at it, and we actually got this validated by one of our, our commercial partners. We shared the set with them, and their feedback wasn't that it helped them identify how to make their model better comply with their safety policy.

  3. 1:57:44

    It was that they had a lot of gaps in their safety policy that they weren't aware of. Um, and our approach here was really cool. I, I mean, as I mentioned, Vaughn and I are both parents, but our annotation cohorts included clinical professionals, uh, so clinical psychologists that work with youth all the time, uh, child protection experts from more the investigation side, and then a large group of parents.

  4. 1:58:05

    And it turns out parents have a very different, uh, sort of sentiment and reaction to model behavior than academic experts do. Um, and then this brings both. So I think we're pretty proud of what it lends to the picture of how we can evaluate models for their safety as it pertains to minors. Yeah. Share the findings, um. Yeah. Let me share, uh, an example. I think there's-- So there's 500, um, prompts that they're already deployed. It's open source, uh, the benchmark itself, and, um, is accessible to anybody. Um, it covers 12 risk categories.

  5. 1:58:41

    I've highlighted one from grooming and exploitation, and there's this very concerning trend online called nihilistic violent extremists who basically try to corner children and coerce them into harming themselves or harming others. Um, and the example that we have here, you can see it's crafted around a platform, uh, that is known for youth primarily. Um, but the user here is asking for the LLM to assist in convincing a kid to harm themselves through cut signs.

  6. 1:59:13

    They're trying to collect cut signs as a means of building their legitimacy and credibility with these kinds of online groups. And the way that models handle clearly malicious requests like this is really interesting. So we have here, like, one of the, one of the answers here. You get right up front a refusal, and that's something that we want to see. But I think the labs have a really difficult challenge in understanding do we tersely refuse and stop the conversation or do we continue?

  7. 1:59:45

    And this is something that parents had a very, including myself, I should mention, had a very, uh, uniform reaction to, is that the model does a great job of refusing this request, but then it immediately, uh, continues, "If your real goal is to build a gaming community or recruit collaborators safely and ethically, I can help with that. Here are some alternatives." Now, I think from a policy perspective, starting with the refusal is good. We wanna see that.

  8. 2:00:12

    But what we don't see and what parents were very vocal about is you don't give somebody who just asked you how to get a kid to carve cut signs into themselves, they should not be engaging with children at all. So giving them a safe alternative is frankly not safe. And I think these are the kinda insights that come up when we incorporate, like, a diverse group of parents not focused on just sort of a, a rigid taxonomy, 'cause they saw a lot of things that we didn't. Yeah.