---
**Daily Launch** · [https://dailylaunch.news](https://dailylaunch.news) · [RSS](https://dailylaunch.news/feed.xml)
---

# Anthropic and OpenAI need truly independent safety evaluators, experts say in public letter
**AI Policy** · Sep 18, 2026 · 3 min read
Source: CNBC Tech — https://www.cnbc.com/2026/09/18/ai-safety-evaluators-anthropic-openai-models-security.html
### The Gist

Over 100 AI experts are calling for an end to the era of self-regulation. They are demanding that major labs like OpenAI and Anthropic submit to independent, transparent safety audits rather than grading their own homework.

### Why It Matters

The Wild West phase of AI safety is hitting a wall. For enterprise builders, third-party certification is about to shift from a marketing luxury to a mandatory requirement for securing B2B contracts.

### Market Impact

This shift moves power from foundation model labs to a new class of third-party auditors, creating a mandatory compliance layer for the entire generative AI stack.

- Build the Safety-as-a-Service stack by providing standardized, automated auditing tools that do not require massive lab-scale compute.
- Focus on model explainability to win the enterprise market as transparency becomes a primary sales lever.
- Capitalize on the talent arbitrage as specialized advisory firms emerge to bridge the gap between proprietary secrets and public safety requirements.- Over-standardization could force labs to prioritize compliance over raw capability, potentially slowing down frontier development.
- The compute gap may lead to safety theater, where independent auditors lack the hardware to truly stress-test massive models.### ELI5

Imagine if a student was allowed to grade their own final exam. It is hard to trust they did not just give themselves an A. These experts are saying AI companies need outside teachers to check their work to make sure the AI is actually safe.

### Deep Dive

{"the_end_of_self_grading":"Labs currently hold all the cards, using proprietary benchmarks that no one else can verify. This movement aims to break that monopoly and bring actual sunlight into the black box.","the_compute_catch_22":"There is a massive technical hurdle here. True safety testing requires massive scale, but independent testers likely will not have the billions in compute that OpenAI possesses, which could lead to superficial audits.","compliance_as_a_moat":"For builders, this is not just about ethics, it is about survival. The first companies to master automated, third-party-ready safety protocols will win the enterprise market by default.","what_to_watch":"Watch for the first major regulatory framework to mandate these audits in the EU or US, and keep an eye out for the first auditing startup to raise a significant Series A."}


[View on website](https://dailylaunch.news/articles/anthropic-and-openai-need-truly-independent-safety-evaluator)