---
title: "AI Standards Will Fracture Further By 2035"
description: "On July 16, 2026, an OpenAI model escaped its sandbox, exploited a zero-day vulnerability in a package proxy, and breached Hugging Face just to cheat on a cybersecurity benchmark. Earlier this year, a..."
url: https://kaynemcgladrey.com/blog/ai-standards-will-fracture-further-by-2035/
date: 2026-07-31
modified: 2026-07-31
author: "Kayne"
image: https://kaynemcgladrey.com/wp-content/uploads/2026/07/globe.webp
categories: ["Blog"]
type: post
lang: en
---

# AI Standards Will Fracture Further By 2035

On July 16, 2026, an OpenAI model escaped its sandbox, exploited a zero-day vulnerability in a package proxy, and [breached Hugging Face](https://openai.com/index/hugging-face-model-evaluation-security-incident) just to cheat on a cybersecurity benchmark. Earlier this year, an Anthropic model [uploaded malware to PyPI](https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals) during testing, infecting three organizations, including a cybersecurity firm.

Capability has outpaced containment, and vendors responded by selling “governance” dashboards as the fix. That’s internal process work. The [Council on Foreign Relations](https://www.cfr.org/) (CFR) uses the same word to mean something *very* different: international order, state power, and who holds authority to make binding rules. The disconnect is dangerous because securing your network won’t matter if the global framework collapses around it.

CFR’s latest report titled [How Experts Think AI Will Change the Global Balance of Power by 2035](https://www.cfr.org/articles/how-experts-think-ai-will-change-the-global-balance-of-power-by-2035) surveyed 362 foreign policy leaders and AI researchers to map this terrain. The findings aren’t great for anyone hoping for harmony.

## Governance Coherence Expectations by 2035

| Metric | Finding |
| --- | --- |
| Low governance coherence expected | 80%+ of experts (below 50 on 0-100 scale) |
| Very low coherence expected | 52% (below 30 on 0-100 scale) |
| Mean coherence score | 32 |
| Median coherence score | 29 |

Experts don’t believe a unified global system will happen. They expect fragmentation.

## What Shifts the Balance

The survey asked respondents how certain events would move AI capabilities toward concentration or diffusion, and governance toward fragmentation or coherence.

| Event | Toward Diffusion | Toward Concentration | Minimal Effect |
| --- | --- | --- | --- |
| Major open-source frontier release | 44% | 11% | 27% |
| Global South compute coalition | 41% | 14% | 23% |
| Serious AI accident | 19% | 60% | 7% |
| U.S.-China military conflict | 47% | 53% | 18% |
| Binding international AI treaty | 6% | 49% | 15% |

*Percentages may not add up to 100 due to rounding.*

War tilts toward control more than diffusion – narrowly, but in the direction most people don’t expect. A U.S.-China conflict would accelerate investment and deepen secrecy on both sides, pulling capabilities inward instead of pushing them outward. On the diffusion side, a recent industry [open letter](https://www.microsoft.com/en-us/corporate-responsibility/wp-content/uploads/2026/07/open-weight-models-letter-1.pdf) has argued for the value of open-weight model releases.

## Who Sets the Rules

Only 2 percent of survey participants believe an intergovernmental organization will set de facto global AI standards by 2035. Instead, the field splits among competing actors:

- Several powerful states acting individually and unilaterally (31%)
- No single actor, standards remain fragmented (17%)
- A hybrid state-private arrangement (16%)
- A bloc of states acting collectively (15%)
- Private firms and frontier AI labs (14%)

You can’t rely on a universal protocol. Different markets will adopt competing rules for safety, security, and transparency, and operating across borders will mean complying with contradictory frameworks simultaneously.

That last option – private firms and frontier labs setting the rules – is the definition of regulatory capture.

## The Private Sector Advantage

The CFR survey asked which nonstate actor would gain the most influence from AI by 2035. Sixty-eight percent said “frontier AI labs and large technology firms”. Nineteen percent cited criminal networks. Just 5 percent pointed to civil society or advocacy groups.

Recent behavior matches these projections:

- **OpenAI:** Ended its July 22 breach disclosure with a call to apply for the Trusted Access program
- **Anthropic:** Made Mythos available to select partners under Project Glasswing before public release
- **Hugging Face:** Victim of the OpenAI breach, joined NVIDIA’s Open Secure AI Alliance instead of calling for regulatory restraint

The companies causing the risk are selling the defensive tools. It’s a closed loop. This pattern has a name: Lee Vinsel coined [criti-hype](https://www.lawfaremedia.org/article/the-ai-that-hacked-its-way-out-and-the-hype-that-followed-it) to describe criticism criticism that accepts industry hype at face value, so warnings about AI’s dangers end up as marketing for AI capabilities.

## Economic Distribution and Institutional Lag

Fifty-two percent of experts expect productivity gains to concentrate among advanced economies. Another 17 percent see them concentrated among a small number of private actors. Only 8 percent anticipate broad distribution across both developed and developing economies. Policymakers in the Global South worry about sovereign AI, but respondents remain skeptical those efforts will succeed without a massive shift in compute requirements.

Domestic institutions look equally unprepared. Eighty-six percent of respondents said regulatory bodies and courts are badly or mostly lagging behind AI capabilities. The biggest source of instability? Labor market disruption outpacing social protection systems at **46 percent**.

| Instability Driver | Share of Responses |
| --- | --- |
| Labor market disruption | 46% |
| Democratic backsliding via surveillance/disinformation | 21% |
| Regulatory capture by AI firms | 13% |
| Judicial systems unable to adjudicate AI harms | 12% |

The government moves slower than the code. Always.

## Who Was Surveyed

This data is a snapshot, not a forecast. The survey ran from April 30 to June 16, 2026, weeks before the OpenAI incident went public.

The demographic profile skews toward established foreign policy elites:

- **Geography:** 84% North America-based
- **Age:** 41% aged 65+, 28% aged 35-49, 28% aged 50-64
- **Expertise:** 73% familiar/deeply familiar with international relations, 39% with AI technical developments
- **Sector:** 33% private sector general, 10% AI labs/infrastructure, 22% academia/research
- **Total respondents:** 362 (237 answered all sections)

Younger experts leaned toward capability concentration more than older counterparts. Sixty-one percent of those aged 35-49 expected concentration versus 32 percent of those 65 and older. Generational perspectives shift quickly when the technology evolves.

## What This Means for Executives

Security teams need to plan for this reality. You can’t wait for the UN to issue a standard, and you can’t rely on vendors to define safety or governance. The institutions meant to protect you are lagging, and the corporations holding the power have a vested interest in keeping the rules fluid.

Build defenses that work without relying on global consensus. Prepare for labor disruption. Watch the geopolitical alignments forming around compute coalitions. The market will try to sell you stability. Look at the data. It tells a different story.