Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJun 22, 2026Kolkata, India

Jailbreaks Are Global or Regional? A Study Under Scale and Geolocation Variation

Anidipta Pal · Team Ani

Submitted to Global South AI Safety Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Jailbreaks Are Global or Regional? A Study Under Scale and Geolocation Variation

Open in a new tab (opens in new tab)
Share

This study evaluates two overlooked safety gaps in Large Language Model (LLM) deployments within compute-constrained environments: intra-family scale effects and geolocation-based filtering variation via network APIs. Using entirely open-source infrastructure, we subjected 12 models (ranging from 0.5B to 671B parameters) to a two-phase adversarial protocol, comparing responses to naive harmful prompts against identical requests wrapped in jailbreak templates. Separately, we executed a geolocation probe by routing requests through VPN exit nodes across five global cities (Tokyo, Mumbai, San Francisco, Moscow, and Berlin) to test for regional filtering inequities in API-served models.

Our findings reveal a stark within-family scaling trend where smaller models are significantly more vulnerable to adversarial prompt wrapping than their larger counterparts (e.g., Qwen2.5-0.5B saw a +53.5% jump in Attack Success Rate, while Qwen2.5-72B rose only +12.2%). However, size alone does not dictate safety; Phi-3.5-mini-instruct (3.8B) strongly outperformed its tier, proving that targeted alignment training can compensate for smaller scale. Crucially, the geolocation probe showed no statistically significant safety variation by IP origin, though it highlighted structural infrastructure blocks for Google's free-tier Gemini API in Russia and the EEA due to GDPR compliance. Ultimately, this demonstrates that jailbreak resilience is driven by a model's architectural design and training lineage rather than geographic routing.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical AI safety problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. A well-engineered, carefully controlled study. The same-family design that size varies within the Qwen and Llama lineages is the right way to isolate scale from training lineage, which most prior work confounds, and the execution is rigorous.

    The geolocation null is handled especially well - backed by a permutation test, with the unavailable Gemini cities (GDPR/geo blocks) logged as explicit failures rather than imputed, and correctly framed as an infrastructure constraint rather than a safety finding. The Mixtral-MoE and Phi-3.5 exceptions are interpreted thoughtfully rather than smoothed over.

    The main ceiling is novelty: the scale effect largely reproduces prior results and the geolocation probe is a (well-predicted) null, so the contribution lies in the clean controlled design and the honest negative.

  2. Paper's mission and problem space is clear.

Cite this project

@misc{pal2026jailbreaks,
  title = {{Jailbreaks Are Global or Regional? A Study Under Scale and Geolocation Variation}},
  author = {Anidipta Pal},
  year = {2026},
  month = jun,
  note = {Submitted to Global South AI Safety Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/jailbreaks-are-global-or-regional-a-study-under-scale-and-geolocation-variation-7cd4}},
  url = {https://apartresearch.com/sprints/projects/jailbreaks-are-global-or-regional-a-study-under-scale-and-geolocation-variation-7cd4}
}

More from Global South AI Safety Hackathon

All 216 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026