---
type: "post"
id: "openai-cannot-make-ai-safe-on-its-own-8154f5"
slug: "openai-cannot-make-ai-safe-on-its-own-8154f5"
title: "OpenAI Cannot Make AI Safe on Its Own (6 minute read)"
description: "**Thesis:** Safety governance depends not only on model safeguards but also on whether researchers can raise concerns and collaborate with independent experts without unclear rules or fear of retaliation."
retrieval_nugget: "**Thesis:** Safety governance depends not only on model safeguards but also on whether researchers can raise concerns and collaborate with independent experts without unclear rules or fear of retaliation."
published_at: "2026-10-10"
updated_at: "2026-10-10"
record_date: "2026-10-10"
date_kind: "published_at"
topics: []
source_urls: ["https://mikitabalesni.com/letter/letter.pdf"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:openai-cannot-make-ai-safe-on-its-own-8154f5"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:b15b032b7bd884bf670d4eab45740d22c795b0c7ac0dfb8c138f8853e4a77ab5","reviewed_at":"2026-10-10T05:05:51.555Z","source_evidence_count":1,"verified_claim_count":2,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":2,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Safety governance depends not only on model safeguards but also on whether researchers can raise concerns and collaborate with independent experts without unclear rules or fear of retaliation.","observed_facts":[{"text":"Three former OpenAI safety researchers—Tomek Korbak, Jasmine Wang, and Mikita Balesni—published a letter addressed to the company’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council.","source_urls":["https://mikitabalesni.com/letter/letter.pdf"]},{"text":"The letter recommends preserving independent evaluator access, maintaining monitorability of frontier models, and clarifying procedures for external collaboration.","source_urls":["https://mikitabalesni.com/letter/letter.pdf"]}],"mechanism":"The authors argue that uncertainty about what conduct is permitted can deter employees from escalating concerns or sharing necessary information with outside evaluators. In that account, clear procedures and protected external collaboration are operational inputs to safety work, not merely employment-policy questions.","why_now":"The letter directly connects the authors’ recent dismissals to their concerns about the future of external safety work and asks OpenAI’s oversight bodies to act. This makes the issue a current governance dispute, rather than a general argument about organizational culture.","implications":["OpenAI’s follow-through on independent evaluator access and published collaboration commitments will be a more observable test than assurances about culture alone.","If the authors’ concern is borne out, fewer candid internal reports and narrower access for external evaluators could weaken scrutiny of frontier-model risks."],"evidence_boundary":"The supplied primary source is the three authors’ letter and supports what they say, recommend, and allege; it does not independently establish the disputed circumstances of their dismissals or the company’s internal policies.","watch_conditions":["Watch for a public response from OpenAI or a concrete update to independent evaluators’ access, scope, or collaboration procedures."],"related_records":[],"new_branch_reason":"No prior New Runtime record is supplied for this specific letter; the supplied archive snippets concern different OpenAI events and do not establish a prior record about this intervention."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/openai-cannot-make-ai-safe-on-its-own-8154f5/","markdown":"https://newruntime.com/posts/openai-cannot-make-ai-safe-on-its-own-8154f5.md","json":"https://newruntime.com/posts/openai-cannot-make-ai-safe-on-its-own-8154f5.json"}
---

# OpenAI Cannot Make AI Safe on Its Own (6 minute read)

## Retrieval answer

**Thesis:** Safety governance depends not only on model safeguards but also on whether researchers can raise concerns and collaborate with independent experts without unclear rules or fear of retaliation.

**Thesis:** Safety governance depends not only on model safeguards but also on whether researchers can raise concerns and collaborate with independent experts without unclear rules or fear of retaliation.

**Observed facts:** Three former OpenAI safety researchers—Tomek Korbak, Jasmine Wang, and Mikita Balesni—published a letter addressed to the company’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council. The letter recommends preserving independent evaluator access, maintaining monitorability of frontier models, and clarifying procedures for external collaboration.

**Mechanism:** The authors argue that uncertainty about what conduct is permitted can deter employees from escalating concerns or sharing necessary information with outside evaluators. In that account, clear procedures and protected external collaboration are operational inputs to safety work, not merely employment-policy questions.

**Why now:** The letter directly connects the authors’ recent dismissals to their concerns about the future of external safety work and asks OpenAI’s oversight bodies to act. This makes the issue a current governance dispute, rather than a general argument about organizational culture.

**Implications:** If the authors’ concern is borne out, fewer candid internal reports and narrower access for external evaluators could weaken scrutiny of frontier-model risks. **First implication:** OpenAI’s follow-through on independent evaluator access and published collaboration commitments will be a more observable test than assurances about culture alone.

**Evidence boundary:** The supplied primary source is the three authors’ letter and supports what they say, recommend, and allege; it does not independently establish the disputed circumstances of their dismissals or the company’s internal policies.

**First watch condition:** Watch for a public response from OpenAI or a concrete update to independent evaluators’ access, scope, or collaboration procedures.

This starts a distinct analytical record because no prior New Runtime record is supplied for this specific letter; the supplied archive snippets concern different OpenAI events and do not establish a prior record about this intervention.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)