The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2512
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T07:45:36.254351Z
status: published
lens: p_space
sector: crime
headline: Human and AI judgements of restorative responses to sexual abuse: The role of perceived severity
dek: Abstract Purpose Restorative justice (RJ) offers an alternative or complement to criminal justice in sexual abuse cases, yet the public remains ambivalent. Given that survivors increasingly turn to large language models (LLMs) for information and support when coping with sensitive justice‐related decisions, the present study examined whether the guidance LLMs provide regarding RJ in sexual abuse cases resembles human views. Methods Participants ( N = 469) were randomly assigned to read a vignette describing intr…
gain_title: (none)
problem_title: In a study comparing 469 people with 120 outputs from Gemini, Grok and DeepSeek, the chatbots gave significantly higher support ratings for restorative justice after sexual abuse than humans did, a gap that matters for survivors who ask AI for guidance.
trace_subject: (none)
gain_reading: (none)
gain_evidence: (none)
problem_reading: In a study comparing 469 people with 120 outputs from Gemini, Grok and DeepSeek, the chatbots gave significantly higher support ratings for restorative justice after sexual abuse than humans did, a gap that matters for survivors who ask AI for guidance.
problem_evidence: LLMs produced significantly higher numeric ratings of RJ support than did humans under the structured prompting conditions used
quick_read: Researchers randomly assigned 469 participants to read a vignette describing sexual abuse within or outside the family, then rate how severe it was and how far they supported restorative justice. They compared those ratings with 120 outputs from Gemini, Grok and DeepSeek. In both abuse contexts, the chatbots gave significantly higher support ratings than people did.

Both people and chatbots judged abuse within the family as more severe than abuse outside it. For people, perceived severity fully explained the link between abuse context and support: the more severe the abuse seemed, the less open they were to restorative justice. For the chatbots the effect was only partial, so severity did not act as a categorical veto.

The authors say this matters because survivors increasingly turn to language models for information and support on justice decisions. They note differences between the three models and call for careful ethical curation of such systems. The results apply to structured prompting conditions.
limitation: The chatbot ratings were generated under structured prompting conditions, and the authors say the higher support held under the conditions used, so results may differ with other prompts or real survivor conversations.
tag: Evidence-backed problem
key_points: The study randomly assigned 469 participants to read a vignette of intrafamilial or extrafamilial sexual abuse. | Responses were compared with 120 total outputs from Gemini, Grok and DeepSeek. | Humans and chatbots both rated intrafamilial abuse as more severe than extrafamilial abuse. | In both contexts, chatbots gave significantly higher numeric ratings of restorative justice support than humans. | Among humans, perceived severity fully mediated the effect of abuse context on support. | Among chatbots, mediation was partial, so severity did not categorically rule out restorative justice. | The three models differed significantly from one another.
rundown: The study addresses restorative justice, which offers an alternative or complement to criminal justice in sexual abuse cases and on which the public remains ambivalent. The authors framed it around survivors who use language models when coping with sensitive justice-related decisions.

Participants rated offence severity and support for restorative justice after reading one vignette. Gemini, Grok and DeepSeek produced the comparison responses, 120 outputs in total, under structured prompting.

In people, higher perceived severity reduced openness to restorative justice and fully accounted for the difference between family and non-family abuse. In the chatbots, severity played a smaller role, and their outputs acknowledged severity while staying relatively open to restorative justice.

The authors describe this as consistent with a cautious, descriptive use of the 'mirror cleaner' metaphor. They say the findings show both potential and a need for careful ethical curation of AI systems used by survivors.
sources:
- peer_reviewed | Legal and Criminological Psychology | https://doi.org/10.1111/lcrp.70053 | 2026-07-22
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

1a3b5b8d069310f67559b014763b69ae95020c8652564bc9d98befb108f7fec7

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000