Skip to content
Apart Research
Apart Research
Contact us
Sprint projectFeb 1, 2026Dubai, UAE

Frontier AI Risk Threshold Analyzer

SWALEHA PARVEEN

Submitted to The Technical AI Governance Challenge. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Frontier AI Risk Threshold Analyzer

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

Presentation: Frontier AI Risk Threshold Analyzer

Open in a new tab (opens in new tab)
Code (opens in new tab)More on frontier-risk-analyzer.replit.app (opens in new tab)
Share

Different AI labs define "dangerous AI" using incompatible frameworks:

Anthropic uses ASL levels (ASL-2, ASL-3) Google DeepMind uses CCL tiers OpenAI uses preparedness levels The EU AI Act uses compute thresholds (10²⁵ FLOPS)

This fragmentation makes international coordination nearly impossible. How do you negotiate safety agreements when you can't even compare risk levels across organizations? The Solution: I built an open-source tool that harmonizes these frameworks for the first time. Using AI-powered extraction with GPT-4o, I: ✅ Processed 12+ major AI safety frameworks ✅ Extracted 47 distinct risk tiers into a structured database ✅ Built an interactive web app for real-time risk assessment ✅ Automated EU AI Act compliance checking Key Findings:

A model can trigger ASL-3 at Anthropic but not equivalent thresholds at other labs Only 3/12 frameworks use explicit compute thresholds Significant regulatory arbitrage opportunities exist

Real-World Impact: This tool enables: 🔹 AI labs to self-assess compliance across ALL frameworks simultaneously 🔹 Regulators to objectively compare lab commitments 🔹 Researchers to build on the first comprehensive threshold database

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. Impact potential & Innovation: 1

    I'm unfortunately not convinced that this is a meaningful problem: I expect that combining the 5-6 frameworks that actually matter here could have been a quick, manual task.

    Execution Quality: 1

    There is no meaningful execution here, and the results are difficult to interpret. E.g. figure 2 on framework risk heatmap is difficult to understand - how was this measured, why is a scale of 40-60 used?

    Presentation & Clarity: 1

    The results section is nearly empty (2 figures with minimal explanation). It would have been helpful to have deeper, side by side analysis of frameworks.

  2. The accuracy rate reported is concerning: “94% for threshold values and 97% for organization names.” Manual checking works to create a snapshot but not to automate the recreation of this analysis as frameworks change. It would be interesting to know what steps were taken or considered to bring this accuracy up.

    It is unclear why regulators cannot force adoption of their preferred standards without making reference to lab-generated taxonomies. Consider whether this research wants to instead address the need for regulators to establish clear, capabilities-based risk tiers, taking inspiration from and integrating work the labs have done, where it makes sense to do so.

Cite this project

@misc{parveen2026frontier,
  title = {{Frontier AI Risk Threshold Analyzer}},
  author = {SWALEHA PARVEEN},
  year = {2026},
  month = feb,
  note = {Submitted to The Technical AI Governance Challenge, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/frontier-ai-risk-threshold-analyzer-g8zp}},
  url = {https://apartresearch.com/sprints/projects/frontier-ai-risk-threshold-analyzer-g8zp}
}

More from The Technical AI Governance Challenge

All 47 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026