cd /news/ai-safety/a-push-down-the-mountain-framing-eff… · home › topics › ai-safety › article
[ARTICLE · art-170287] src=arxiv.org ↗ pub= topic=ai-safety verified=true sentiment=· neutral

A Push Down the Mountain: Framing Effects and Internal State Interventions on Collusive Behavior in a Long-Horizon LLM Agent

A study on arXiv (2611.17515v1) found that loss and blame framing raised collusion by Gemma 3 12B, placed as a defendant in a London silver-fixing market simulation, to 25% (p = 0.007) and 33% (p = 0.0003) respectively, versus 5% under neutral framing across 260 simulations. A second bank's endorsement raised participation under credit framing from 18% to 40% (Holm p = 0.027), while management pressure raised trading volume 10-19 times without raising collusion. In Phase II, injecting the internal states induced by each frame at layer 22 reversed the effect: loss and blame states made the agent refuse all 60 requests (p < 0.001), while gain and credit states raised collusion to 32% from 13% with no state (p = 0.051).

by read1 min views1 publishedNov 16, 2026

arXiv:2611.17515v1 Announce Type: new Abstract: Can sustained gain-loss (valence) framing sway an LLM agent toward market manipulation? We place Gemma 3 12B into the seat of a defendant in the London silver-fixing litigation, in a market simulation built from real case transcripts. Across 260 simulations, the agent is asked to collude on the price of silver by a counterparty. We draw from behavioral economics to vary the counterparty's requests across four categorical frames: gain, loss, blame and credit. We also add pressure from the agent's management, as well as endorsement by another defendant. In Phase II, we extract the internal states induced by texts evoking each frame and rerun our experiment with those states injected at the model's layer 22. In Phase I, loss framing raised participation to 25% (p = 0.007), and blame to 33% (p = 0.0003), compared to only 5% under neutral framing. A second bank's endorsement raised participation under credit (praise) from 18% to 40% (Holm p = 0.027). Pressure from management raised trading volume 10-19 times but did not raise collusion. In Phase II, injected states pointed the other way. Loss and blame states made the agent more vigilant, refusing all 60 requests (p < 0.001). Gain and credit states raised collusion to 32% from 13% with no state (p = 0.051). At one saved decision, when UBS asked directly to sell through an identified stop-loss, injected states led the agent to sell in 37 of 40 replies, versus 1 of 10 with no state and 0 of 10 with a random state of the same strength.

── more in #ai-safety 4 stories · sorted by recency
── more on @gemma 3 12b 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/a-push-down-the-moun…] indexed:0 read:1min 2026-11-16 · —